This PDF file contains the front matter associated with SPIE Proceedings Volume 7537, including the Title Page, Copyright information, Table of Contents, the Conference Committee listing and introduction.

We are just a few years away from celebrating the 200^th anniversary of photography. The first permanent photographic record was made by Niepce in 1826, the view from his window at Le Gras. After many development cycles, including some periods of stagnation, photography is now experience an amazing period of growth. Change since the mid 90's going into the next several years will completely modify photography and its industry. We propose that the digital photography revolution can be divided into two phases. The first, from about 1994 to 2009, was primarily the transformation of film-based equipment into their digital counterparts. Now, in the second phase, photography is starting to change into something completely different, with forces like social networks, cell phone cameras and computational photography changing the business, the methods and the use of photographs.

In this study, we evaluated the effect that pixel size has upon people's preferences for images. We used multispectral images of faces as the scene data and simulated the response from sensors with different pixel size while the other sensor parameters were kept constant. Subjects were asked to choose between pairs of images; we found that preference judgments were primarily influenced by the visibility of uncorrelated noise in the images. We used the SCIELAB metric (ΔE) to predict the visibility of the uncorrelated image noise. The S-CIELAB difference between a test image and an ideal reference image was monotonically related to the preference score.

Reducing hot pixels is a challenge commonly faced in the image sensor industry and there are various techniques used to address this problem, including image processing and process optimization. This paper discusses an approach to reduce hot pixels by using Technology Computer Aided Design (TCAD) simulations to optimize the pixel at the process level. A correlation between empirical hot pixel data and simulated electric field is discussed. For this given process, there is good correlation between hot pixel count and the electric field along the top p-n junction of the photodiode. By optimizing the top p-n junction, we were able to reduce the hot pixel count to less than 100ppm at 45C for a threshold value of 15% of full scale. However, careful consideration must be made during the process optimization. When photodiode implant doses and energies are changed, image lag performance can deteriorate. Changing photodiode implant doses and energies can also result in n-type penetration through the polysilicon gate, which can lead to increased dark current. A careful design will avoid such problems. During our process optimization, we successfully reduced hot pixel count while still achieving low dark current. These achievements can be observed in dark current of less than 3 e- /sec-pixel at 45C.

Accurate noise level estimation is essential to assure good performance of noise reduction filters. Noise contaminating raw images is typically modeled as additive white and Gaussian distributed (AWGN); however raw images are affected by a mixture of noise sources that overlap according to a signal dependent noise model. Hence, the assumption of constant noise level through all the dynamic range represents a simplification that does not allow precise sensor noise characterization and filtering; consequently, local noise standard deviation depends on signal levels measured at each location of the CFA (Color Filter Array) image. This work proposes a method for determining the noise curves that map each CFA signal intensity to its corresponding noise level, without the need of a controlled test environment and specific test patterns. The process consists in analyzing sets of heterogeneous raw CFA images, allowing noise characterization of any image sensor. In addition we show how the estimated noise level curves can be exploited to filter a CFA image, using an adaptive signal dependent Gaussian filter.

Threshold estimation in various image noise suppression and edge sharpening algorithms may still contain traces of subjective determination, with groups of subjects preferring more or less than optimal amounts of noise reduction and respectively edge enhancement. We propose in this paper a novel image noise reduction and edge enhancement technique that includes a profiling step for optimal noise threshold selection and resulting spatial frequency response (SFR). Our filter is implemented in the wavelet domain due to flexibility of spatially examining frequencies of interest. The method allows for selectively steering the algorithm to filter the low light 'chrominance noise' of certain hues in color filter array (CFA) cameras more so than others, and for building of a camera noise 'pseudo profile'.

Four channel sensors were evaluated with a image sensor model and their performance was compared with three channel sensors considering both color reproduction accuracy and photo shot noise. When noise was not considered, a sensor with usual RGB plus an additional channel which resided between G and B was the best. But when the emphasis was on noise, a sensor with B, R and two Gs was the best because reducing noise of G should be effective in reducing noise of all L*a*b* components. Bayer color filter array (CFA) samples twice density of G than R or B. This CFA is considered to be efficient in resolution but the result suggests it is also efficient in SNR. Comparing to three channel sensors, the four channel sensors were better in color reproduction but worse in noise. An image preference model proposed by Kuniba and Berns was used to evaluate them and it was shown that neither one was superior than the other.

The current approach used for demosaic algorithm evaluation is mostly empirical and does not offer a meaningful quantitative metric - this disconnects the theoretical results from the results seen in practice. In camera phones, the difference is even bigger due to the low signal to noise ratios and also due to the overlapping of the color filters. This implies that a demosaic algorithm has to be designed to allow for graceful degradation in presence of noise. Also, the demosaic algorithm has to be tolerant to high color correlations. In this paper we propose a special class of images and a methodology that can be used to produce a metric indicative of a real case demosaic algorithm performance. The test image that we propose is formed by using a dual chirp signal that is a function of the distance from the center.

As white balance algorithms employed in mobile phone cameras become increasingly sophisticated by using, e.g., elaborate white-point estimation methods, a proper color calibration is necessary. Without such a calibration, the estimation of the light source for a given situation may go wrong, giving rise to large color errors. At the same time, the demands for efficiency in the production environment require the calibration to be as simple as possible. Thus it is important to find the correct balance between image quality and production efficiency requirements. The purpose of this work is to investigate camera color variations using a simple model where the sensor and IR filter are specified in detail. As input to the model, spectral data of the 24-color Macbeth Colorchecker was used. This data was combined with the spectral irradiance of mainly three different light sources: CIE A, D65 and F11. The sensor variations were determined from a very large population from which 6 corner samples were picked out for further analysis. Furthermore, a set of 100 IR filters were picked out and measured. The resulting images generated by the model were then analyzed in the CIELAB space and color errors were calculated using the ΔE94 metric. The results of the analysis show that the maximum deviations from the typical values are small enough to suggest that a white balance calibration is sufficient. Furthermore, it is also demonstrated that the color temperature dependence is small enough to justify the use of only one light source in a production environment.

Recently, R. H. Chan, T. F. Chan, L. Shen, and Z. Shen proposed an image-restoration method, referred to as the C²-S² method, in the shift-invariant Haar wavelet transform domain. The C²-S² method is suitable for image-restoration in a digital color camera and restores a sharp color image in a lightly noisy case; but in a heavily noisy case it produces colored artifacts originating from noise. In such a case, the image-restoration process should be split into a colorinterpolation stage, a denoising stage, and a deblurring stage; and along this line we present an approach to restore a high ISO-sensitivity color image. Our approach firstly de-mosaics observed color data with the bi-linear interpolation method, which is robust against observation noise but causes image blurs. Next, our approach applies our previouslyproposed spatially-adaptive soft color-shrinkage to each shift-invariant Haar wavelet coefficient of the demosaicked color-image, to produce a denoised color-image. Finally, our approach applies to the denoised color-image a colorimage restoration method that corresponds to an extension of the C²-S² method and employs our previously proposed soft color-shrinkage. The experimental simulations conducted on raw color data observed with a SLR digital color camera with 6400 ISO-sensitivity demonstrate that our approach restores a high-quality color image even in a high ISOsensitivity case, without producing noticeable artifacts.

Conventional point spread function (PSF) measurement methods often use parametric models for the estimation of the PSF. This limits the shape of the PSF to a specific form provided by the model. However, there are unconventional imaging systems like multispectral cameras with optical bandpass filters, which produce an, e.g., unsymmetric PSF. To estimate such PSFs we have developed a new measurement method utilizing a random noise test target with markers: After acquisition of this target, a synthetic prototype of the test target is geometrically transformed to match the acquired image with respect to its geometric alignment. This allows us to estimate the PSF by direct comparison between prototype and image. The noise target allows us to evaluate all frequencies due to the approximately "white" spectrum of the test target - we are not limited to a specifically shaped PSF. The registration of the prototype pattern gives us the opportunity to take the specific spectrum into account and not just a "white" spectrum, which might be a weak assumption in small image regions. Based on the PSF measurement, we perform a deconvolution. We present comprehensive results for the PSF estimation using our multispectral camera and provide deconvolution results.

We introduce a new metric, the visible signal-to-noise ratio (vSNR), to analyze how pixel-binning and resizing methods influence noise visibility in uniform areas of an image. The vSNR is the inverse of the standard deviation of the SCIELAB representation of a uniform field; its units are 1/ΔE. The vSNR metric can be used in simulations to predict how imaging system components affect noise visibility. We use simulations to evaluate two image rendering methods: pixel binning and digital resizing. We show that vSNR increases with scene luminance, pixel size and viewing distance and decreases with read noise. Under low illumination conditions and for pixels with relatively high read noise, images generated with the binning method have less noise (high vSNR) than resized images. The binning method has noticeably lower spatial resolution. The binning method reduces demands on the ADC rate and channel throughput. When comparing binning and resizing, there is an image quality tradeoff between noise and blur. Depending on the application users may prefer one error over another.

The dead leaves model was recently introduced as a method for measuring the spatial frequency response (SFR) of camera systems. The target consists of a series of overlapping opaque circles with a uniform gray level distribution and radii distributed as r^-3. Unlike the traditional knife-edge target, the SFR derived from the dead leaves target will be penalized for systems that employ aggressive noise reduction. Initial studies have shown that the dead leaves SFR correlates well with sharpness/texture blur preference, and thus the target can potentially be used as a surrogate for more expensive subjective image quality evaluations. In this paper, the dead leaves target is analyzed for measurement of camera system spatial frequency response. It was determined that the power spectral density (PSD) of the ideal dead leaves target does not exhibit simple power law dependence, and scale invariance is only loosely obeyed. An extension to the ideal dead leaves PSD model is proposed, including a correction term to account for system noise. With this extended model, the SFR of several camera systems with a variety of formats was measured, ranging from 3 to 10 megapixels; the effects of handshake motion blur are also analyzed via the dead leaves target.

We describe the procedure to evaluate the image quality of a camera in terms of texture preservation. We use a stochastic model coming from stochastic geometry, and known as the dead leaves model. It intrinsically reproduces occlusions phenomena, producing edges at any scale and any orientation with a possibly low level of contrast. An advantage of this synthetic model is that it provides a ground truth in terms of image statistics. In particular, its power spectrum is a power law, as many natural textures. Therefore, we can define a texture MTF as the ratio of the Fourier transform of the camera picture by the Fourier transform of the original target and we fully describe the procedure to compute it. We will compare the results with the traditional MTF (computed on a slanted edge as defined in the ISO 12233 standard) and will show that the texture MTF is indeed more appropriate for describing fine detail rendering.

The aim of the paper is to define an objective measurement for evaluating the performance of a digital camera. The challenge is to mix different flaws involving geometry (as distortion or lateral chromatic aberrations), light (as luminance and color shading), or statistical phenomena (as noise). We introduce the concept of information capacity that accounts for all the main defects than can be observed in digital images, and that can be due either to the optics or to the sensor. The information capacity describes the potential of the camera to produce good images. In particular, digital processing can correct some flaws (like distortion). Our definition of information takes possible correction into account and the fact that processing can neither retrieve lost information nor create some. This paper extends some of our previous work where the information capacity was only defined for RAW sensors. The concept is extended for cameras with optical defects as distortion, lateral and longitudinal chromatic aberration or lens shading.

Recently, we have proposed a new image device called gigavision camera whose most important characteristic is that pixels have binary response. The response function of a gigavision sensor is non-linear and similar to a logarithmic function, which makes the camera suitable for high dynamic range imaging. One important parameter in the gigavision camera is the threshold for generating binary pixels. Threshold T relates to the number of photo-electrons necessary for the pixel output to switch from "0" to "1". In this paper, a theoretical analysis of the threshold influence in the gigavision camera is given. If the threshold in the gigavision sensor is large, there will be a "dead zone" in the response function of a gigavision sensor. A method of adding artificial light is proposed to solve the "dead zone" problem. Through theoretical analysis and experimental results based on synthesized images, we show that for high light intensity, the gigavision camera with a large threshold and added light works better than one with unity threshold. Experimental results with a prototype camera based on a single photon avalanche diodes (SPAD) camera are also presented.

Defocus imaging techniques, involving the capture and reconstruction of purposely out-of-focus images, have recently become feasible due to advances in deconvolution methods. This paper evaluates the feasibility of defocus imaging as a means of increasing the effective dynamic range of conventional image sensors. Blurring operations spread the energy of each pixel over the surrounding neighborhood; bright regions transfer energy to nearby dark regions, reducing dynamic range. However, there is a trade-off between image quality and dynamic range inherent in all conventional sensors. The approach involves optically blurring the captured image by turning the lens out of focus, modifying that blurred image with a filter inserted into the optical path, then recovering the desired image by deconvolution. We analyze the properties of the setup to determine whether any combination can produce a dynamic range reduction with acceptable image quality. Our analysis considers both properties of the filter to measure local contrast reduction, as well as the distribution of image intensity at different scales as a measure of global contrast reduction. Our results show that while combining state-of-the-art aperture filters and deconvolution methods can reduce the dynamic range of the defocused image, providing higher image quality than previous methods, rarely does the loss in image fidelity justify the improvements in dynamic range.

Recent research demonstrates the advantage of designing electro-optical imaging systems by jointly optimizing the optical and digital subsystems. The optical systems designed using this joint approach intentionally introduce large and often space-varying optical aberrations that produce blurry optical images. Digital sharpening restores reduced contrast due to these intentional optical aberrations. Computational imaging systems designed in this fashion have several advantages including extended depth-of-field, lower system costs, and improved low-light performance. Currently, most consumer imaging systems lack the necessary computational resources to compensate for these optical systems with large aberrations in the digital processor. Hence, the exploitation of the advantages of the jointly designed computational imaging system requires low-complexity algorithms enabling space-varying sharpening. In this paper, we describe a low-cost algorithmic framework and associated hardware enabling the space-varying finite impulse response (FIR) sharpening required to restore largely aberrated optical images. Our framework leverages the space-varying properties of optical images formed using rotationally-symmetric optical lens elements. First, we describe an approach to leverage the rotational symmetry of the point spread function (PSF) about the optical axis allowing computational savings. Second, we employ a specially designed bank of sharpening filters tuned to the specific radial variation common to optical aberrations. We evaluate the computational efficiency and image quality achieved by using this low-cost space-varying FIR filter architecture.

The under constrained nature of illuminant estimation determines that in order to resolve the problem, certain assumptions are needed, such as the gray world theory. Including more constraints in this process may help explore the useful information in an image and improve the accuracy of the estimated illuminant, providing that the constraints hold. Based on the observation that most personal images have contents of one or more of the following categories: neutral objects, human beings, sky, and plants, we propose a method for illuminant estimation through the clustering of pixels of gray and three dominant memory colors: skin tone, sky blue, and foliage green. Analysis shows that samples of the above colors cluster around small areas under different illuminants and their characteristics can be used to effectively detect pixels falling into each of the categories. The algorithm requires the knowledge of the spectral sensitivity response of the camera, and a spectral database consisted of the CIE standard illuminants and reflectance or radiance database of samples of the above colors.

Face detection has been implemented in many digital still cameras and camera phones with the promise of enhancing existing camera functions (e.g. auto exposure) and adding new features to cameras (e.g. blink detection). In this study we examined the use of face detection algorithms in assisting auto exposure (AE). The set of 706 images, used in this study, was captured using Canon Digital Single Lens Reflex cameras and subsequently processed with an image processing pipeline. A psychophysical study was performed to obtain optimal exposure along with the upper and lower bounds of exposure for all 706 images. Three methods of marking faces were utilized: manual marking, face detection algorithm A (FD-A), and face detection algorithm B (FD-B). The manual marking method found 751 faces in 426 images, which served as the ground-truth for face regions of interest. The remaining images do not have any faces or the faces are too small to be considered detectable. The two face detection algorithms are different in resource requirements and in performance. FD-A uses less memory and gate counts compared to FD-B, but FD-B detects more faces and has less false positives. A face detection assisted auto exposure algorithm was developed and tested against the evaluation results from the psychophysical study. The AE test results showed noticeable improvement when faces were detected and used in auto exposure. However, the presence of false positives would negatively impact the added benefit.

The high level context image analysis regards many fields as face recognition, smile detection, automatic red eye removal, iris recognition, fingerprint verification, etc. Techniques involved in these fields need to be supported by more powerful and accurate routines. The aim of the proposed algorithm is to detect elliptical shapes from digital input images. It can be successfully applied in topics as signal detection or red eye removal, where the elliptical shape degree assessment can improve performances. The method has been designed to handle low resolution and partial occlusions. The algorithm is based on the signature contour analysis and exploits some geometrical properties of elliptical points. The proposed method is structured in two parts: firstly, the best ellipse which approximates the object shape is estimated; then, through the analysis and the comparison between the reference ellipse signature and the object signature, the algorithm establishes if the object is elliptical or not. The first part is based on symmetrical properties of the points belonging to the ellipse, while the second part is based on the signature operator which is a functional representation of a contour. A set of real images has been tested and results point out the effectiveness of the algorithm in terms of accuracy and in terms of execution time.

The aim of this research is to derive illuminant-independent type of HDR imaging modules which can optimally multispectrally reconstruct of every color concerned in high-dynamic-range of original images for preferable cross-media color reproduction applications. Each module, based on either of broadband and multispectral approach, would be incorporated models of perceptual HDR tone-mapping, device characterization. In this study, an xvYCC format of HDR digital camera was used to capture HDR scene images for test. A tone-mapping module was derived based on a multiscale representation of the human visual system and used equations similar to a photoreceptor adaptation equation, proposed by Michaelis-Menten. Additionally, an adaptive bilateral type of gamut mapping algorithm, using approach of a multiple conversing-points (previously derived), was incorporated with or without adaptive Un-sharp Masking (USM) to carry out the optimization of HDR image rendering. An LCD with standard color space of Adobe RGB (D65) was used as a soft-proofing platform to display/represent HDR original RGB images, and also evaluate both renditionquality and prediction-performance of modules derived. Also, another LCD with standard color space of sRGB was used to test gamut-mapping algorithms, used to be integrated with tone-mapping module derived.

Motion due to digital camera movement during the image capture process is a major factor that degrades the quality of images and many methods for camera motion removal have been developed. Central to all techniques is the correct recovery of what is known as the Point Spread Function (PSF). A very popular technique to estimate the PSF relies on using a pair of gyroscopic sensors to measure the hand motion. However, the errors caused either by the loss of the translational component of the movement or due to the lack of precision in gyro-sensors measurements impede the achievement of a good quality restored image. In order to compensate for this, we propose a method that begins with an estimation of the PSF obtained from 2 gyro sensors and uses a pair of under-exposed image together with the blurred image to adaptively improve it. The luminance of the under-exposed image is equalized with that of the blurred image. An initial estimation of the PSF is generated from the output signal of 2 gyro sensors. The PSF coefficients are updated using 2D-Least Mean Square (LMS) algorithms with a coarse-to-fine approach on a grid of points selected from both images. This refined PSF is used to process the blurred image using known deblurring methods. Our results show that the proposed method leads to superior PSF support and coefficient estimation. Also the quality of the restored image is improved compared to 2 gyro only approach or to blind image de-convolution results.

The paper relates to a method for effective reduction of artifacts, caused by lossy compression algorithms based on block-based discreet cosine transform (DCT) coding, known as JPEG coding. Most common artifacts produced by such type of coding, are blocking and ringing artifacts. To reduce the effect of coding artifacts caused by significant information loss, a variety of different algorithms and methods has been suggested. However, the majority of solutions propose to process all blocks in the image, which leads to increase of processing time, required resources, as well as image over-blurring after processing of blocks, not affected by blocking artifacts. Techniques for ringing artifact detection usually rely on edge-detection step, a complicated and versatile procedure with unknown optimal parameters. In this paper we describe very effective procedures for detection of artifacts, and their subsequent correction. This approach helps to save notable amount of computational resources, since not all the blocks are involved in correction procedures. Detection steps are performed in frequency domain, using only DCT coefficients of an image. Numerous examples have been analyzed and compared with the existent solutions, and results prove the effectiveness of proposed technique.

The pixel densities of current CMOS sensors increase and bring new challenges for image sensor designers. Todays sensor modules with miniature lenses often exhibit a considerable amount of color lens shading. This shading is spatial-variant and can be easily identified when capturing a flat textureless Lambertian surface and inspecting the light fall-off and hue change from the image center towards the borders. In this paper we discuss lens shade compensation using spatially dependent gains for each of the four color channel in the Bayer color filter array. We determine reference compensation functions in off-line calibration and efficiently parameterize each function with a bilinear spline which we fit to the reference function using constrained least-squares and Lagrangian conditions ensuring continuity between the piece-wise bilinear functions. For each spline function we optimize a rectilinear grid on which the spline knots are aligned by minimizing the square errors between reference and approximated compensation function. Our evaluations provide quantitative results with real image data using three recent CMOS sensor modules.

The classic shrinkage works well for monochrome-image denoising. To utilize inter-channel color correlations, a noisy image undergoes the color-transformation from the RGB to the luminance-and-chrominance color space, and the luminance and the chrominance components are separately denoised. However, this approach cannot cope with signaldependent noise of a digital color camera. To utilize the noise's signal-dependencies, previously we have proposed the soft color-shrinkage where the inter-channel color correlations are directly utilized in the RGB color space. The soft color-shrinkage works well; but involves a large amount of computations. To alleviate the drawback, taking up the l⁰-l² optimization problem whose solution yields the hard shrinkage, we introduce the l0 norms of color differences and the l⁰ norms of color sums into the model, and derive hard color-shrinkage as its solution. For each triplet of three primary colors, the hard color-shrinkage has 24 feasible solutions, and from among them selects the optimal feasible solution giving the minimal energy. We propose a method to control its shrinkage parameters spatially-adaptively according to both the local image statistics and the noise's signal-dependencies, and apply the spatially-adaptive hard color-shrinkage to removal of signal-dependent noise in a shift-invariant wavelet transform domain. The hard color-shrinkage performs mostly better than the soft color-shrinkage, from objective and subjective viewpoints.

Correct image orientation is often assumed by common imaging applications such as enhancement, browsing, and retrieval. However, the information provided by camera metadata is often missing or incorrect. In these cases manual correction is required, otherwise the images cannot be correctly processed and displayed. In this work we propose a system which automatically detects the correct orientation of digital photographs. The system exploits the information provided by a face detector and a set of low-level features related to distributions in the image of color and edges. To prove the effectiveness of the proposed approach we evaluated it on two datasets of consumer photographs.

We have designed a new self-adaptive image cropping algorithm that is able to detect several relevant regions in the image. These regions can then be sequentially proposed as thumbnails, to the user according to their relevance order, thus allowing the viewer to visualize the relevant image content and eventually to display or print only those regions in which he is more interested in. The algorithm exploits both visual and semantic information. Visual information is obtained by a visual attention model, while semantic information relates to the detection and recognition of particularly significant objects. In this work we concentrate our attention on the two common objects found in personal photos, such as face and skin regions. Examples are shown to illustrate the effectiveness of the proposed method.

Many times when digital cameras are tested time is short. May it be because the test is performed on a production line where productivity is one of the most important aspects or in a lab where a schedule needs to be met. To reduce testing time one of the important aspects is to reduce the amount of images that need to be taken and analyzed. This can be achieved by combining different features into a single test target. Another reason to use combined targets is to eliminate any shot to shot variations in exposure, focus setting, or image processing. Last but not least a single target may be cheaper than buying several targets for the different image quality aspects. But when different features are incorporated into a single chart a couple of questions arise. Can the different aspects be determined independently? Is it better to use a transparent or a reflective target? Can e.g. a reflective chart with its limited contrast be used to measure the dynamic range of a camera? This paper discusses the pros and cons of combined test charts. It describes the related possibilities and problems for the main aspects of image quality like OECF, resolution, color reproduction, shading, and distortion. Application specific possibilities are studied and summarized.

Market's demands of digital cameras for higher sensitivity capability under low-light conditions are remarkably increasing nowadays. The digital camera market is now a tough race for providing higher ISO capability. In this paper, we explore an approach for increasing maximum ISO capability of digital cameras without changing any structure of an image sensor or CFA. Our method is directly applied to the raw Bayer pattern CFA image to avoid non-linearity characteristics and noise amplification which are usually deteriorated after ISP (Image Signal Processor) of digital cameras. The proposed method fuses multiple short exposed images which are noisy, but less blurred. Our approach is designed to avoid the ghost artifact caused by hand-shaking and object motion. In order to achieve a desired ISO image quality, both low frequency chromatic noise and fine-grain noise that usually appear in high ISO images are removed and then we modify the different layers which are created by a two-scale non-linear decomposition of an image. Once our approach is performed on an input Bayer pattern CFA image, the resultant Bayer image is further processed by ISP to obtain a fully processed RGB image. The performance of our proposed approach is evaluated by comparing SNR (Signal to Noise Ratio), MTF50 (Modulation Transfer Function), color error ∝E*ab and visual quality with reference images whose exposure times are properly extended into a variety of target sensitivity.

We propose a method for measuring the multidirectional modulation transfer function (MTF) of digital image acquisition devices using a Siemens star. There are two leading methods for measuring the MTF: the slanted-edge method and the modulated Siemens star method. The former measures the MTF in the horizontal or vertical spatial frequency based on the line spread function (LSF) derived from the edge profile of a slanted knife-edge image. The latter measures the multidirectional MTF using a pattern circumferentially modulated with continuous gray levels. Our method measures the multidirectional MTF using the multidirectional knife-edges of a Siemens star, which is a simple binary image consisting of radial spokes. The vertical edge of the Siemens star is slightly slanted so that the multidirectional edge profiles are obtained in super-resolution. A portion image consisting of the knife-edge is selected in each direction and rotated so that the knife-edge stands upright with a slight tilt. Along the edge slope detected by fitting a cumulative distribution function to the pixel levels, the pixels are projected onto the horizontal axis, forming the edge profile. The resulting multidirectional MTF computed from the edge profiles is in excellent agreement with that measured by the modulated Siemens star method.

Most contemporary still and video cameras employ various optically birefringent materials as optical low-pass filters (OLPF) in order to minimize alias artifacts in the image. Due to the slope characteristics of these filters, camera designers are faced with the choice of either under correcting to maintain image resolution, allowing some aliasing, or eliminating aliases with a more aggressive design that will also compromise the image modulation transfer function (MTF). Furthermore, most OLPFs are designed as optical elements with a frequency response that does not change even if the frequency responses of other elements of the capturing systems are altered. In this paper, we demonstrate the use of a parallel optical window or, alternatively, a rigid mirror positioned between a lens and an imager as an OLPF. Controlled x- and y-axis rotations of the window and the mirror result in a manipulation of the point spread function (PSF) of the system and frequency content of imaged continuous scenes. We evaluate the system MTF when various window functions are used to shape PSF, such as rectangle, triangle, Blackman-Harris etc. We also present results of our experiments when the dynamic OLPF's support and shape are altered to accommodate the optical performance of lenses and imager characteristics.

A careful mathematical analysis of the meaning of variables and equations used in standards for exposure meters and the determination of sensitivities S demonstrates that standards and authors of many photographic texts have erred in interpretations and applications of the common exposure equation. This article concludes that it is inappropriate to use the exposure meter constant K_s as an exposure meter calibration constant because it is essentially a label for a product of characteristics of the photosensitive array employed (the reference exposure H_o = Hsp/_S and the midtone shift M = H_mid/H_sp). It also concludes that the sensitivity and the common exposure equation ultimately depend on the midtone photosensitive exposure target H_mid. This midtone exposure equation can be generalized by including a shift to a arbitrary (non-midtone) photosensitive exposure target in its derivation. This more general exposure equation includes the exposure compensation and eliminates the need for exposure indices. Analysis of the exposure equation for incidentlight exposure meters shows that these meters avoid the vagaries of the current scene by calculating exposure for a potentially very different standard scene and often can be, in effect, less accurate in exposure calculations than reflectedlight exposure meters.

The evaluation of CMOS sensors performance in terms of color accuracy and noise is a big challenge for camera phone manufacturers. On this paper, we present a tool developed with Matlab at STMicroelectronics which allows quality parameters to be evaluated on simulated images. These images are computed based on measured or predicted Quantum Efficiency (QE) curves and noise model. By setting the parameters of integration time and illumination, the tool optimizes the color correction matrix (CCM) and calculates the color error, color saturation and signal-to-noise ratio (SNR). After this color correction optimization step, a Graphics User Interface (GUI) has been designed to display a simulated image at a chosen illumination level, with all the characteristics of a real image taken by the sensor with the previous color correction. Simulated images can be a synthetic Macbeth ColorChecker, for which reflectance of each patch is known, or a multi-spectral image, described by the reflectance spectrum of each pixel or an image taken at high-light level. A validation of the results has been performed with ST under development sensors. Finally we present two applications one based on the trade-offs between color saturation and noise by optimizing the CCM and the other based on demosaicking SNR trade-offs.