Quantitative models are used to predict printer output colors from a set of toner control values. The so-called Neugebauer model predicts the spectral reflectance of the printed colors. We present an algorithm for estimating the parameters of the Neugebauer model from a set of measured reflectances. The algorithm requires a bound on the magnitude of the output reflectance errors. A case study using a high-end color printer suggests that this algorithm yields robust spectral models, which are less sensitive (to undesirable variations in the data) than the models obtained with the methods of least-squares and total least-squares.

The color image has different edge profiles depending on the objects placed in the scene. We propose a novel image-sharpening method adaptive to the local edge slopes with the suppression of background noises. Prescanning the image by a Gaussian derivative (GD) filter, we generate the edge map, which classifies the edge areas to hard, medium, and soft edges, and separates the flat areas without edges. Multiple GD filters with different standard deviations are selectively applied to sharpen each segmented edge area by looking up the edge map. To keep the gray balance, the edge-sharpening filters are applied only to the luminance image. In flat areas except edges, the sharpening filters are resumed and instead, a Gaussian smoothing filter is applied to reduce the background noises. The proposed method brings a dramatic improvement in the reduction of flat-area noises and the natural image-sharpening effects adaptive to the edge slopes. In addition, we newly introduce quality assessment indices to evaluate the image sharpness and flat-area noises with the experimental data.

New applications currently demand utilizing computed tomography (CT) scout images for diagnostic purposes. However, many CT scout images cannot be used diagnostically due to their poor resolution, particularly in the direction of table movement, and loss of detail when displayed with one view. We present two methods to address these two problems. First, spatial resolution generally can be improved with image restoration techniques. Based on the principles of Wiener filtering and inverse filtering, a modified Wiener filtering approach is presented in the frequency domain. The concept of an equivalent target point spread function is also introduced, which makes the restoration process steerable. Consequently, balancing resolution improvement with noise suppression is facilitated. Relevant experiments compare the image quality with traditional inverse filtering and Wiener filtering. The modified Wiener filtering method has been shown to restore the scout image with higher resolution and lower noise. In addition, CT scout images have a wide dynamic range, from 0 to 105 intensity values. They are difficult to display in full detail with only 8 bits (256 intensities). An image fusion approach is developed to preserve and enhance details of CT scout images. The enhanced image is obtained by high-boosting one fused image from another, both of which are computed by fusing a set of preenhanced subimages which derived from the original, using different fusion rules. Image fusion is performed pixel by pixel by the discrete wavelet transform. Final experiments compare the image quality of the resolution-improved and detail-enhanced image and the noise level with those of the original image. Results show that more details, easily observed by a radiologist, are present in the restored and enhanced image than in the original.

We present a regularized mixed norm multichannel image restoration algorithm. The problem of multichannel restoration using both within- and between-channel deterministic information is considered. For each channel a functional that combines the least mean squares (LMS), the least mean fourth (LMF), and a smoothing functional is proposed. We introduce a mixed norm parameter that controls the relative contribution between the LMS and the LMF, and a regularization parameter that defines the degree of smoothness of the solution, both updated at each iteration according to the noise characteristics of each channel. The novelty of the proposed algorithm is that no knowledge of the noise distribution for each channel is required, and the parameters just mentioned are adjusted based on the partially restored image.

We investigate the characteristics of compression noise in images compressed by scalar quantization of the data's wavelet transform coefficients. Such quantization noise is both experimentally and theoretically shown to be spatially varying in the pixel domain, with statistical correlations between the errors at the pixel locations. A quantization covariance matrix is presented that can find use in general restoration scenarios where the observed image or images have been compressed by scalar quantization of image wavelet coefficients. Deblurring is presented as an example use of the quantization model, which demonstrates the model's advantage over the common assumption of independent and identically distributed noise.

A study of a class of openings and closings is investigated using reconstruction criteria. The main goal in studying these transformations consists of eliminating some inconveniences of the morphological opening (closing) and the opening (closing) by reconstruction. The idea in building these new openings and closings comes from the notions of filters by reconstruction and levelings. In particular, concerning the notion of levelings, a study of a class of lower and upper levelings is carried out. The original work of levelings is due to Meyer, who proposes this notion and introduces some criteria to build the levelings in the general case (extended levelings and self-dual transformations). We see the criteria proposed by Meyer as reconstruction criteria during the reconstruction process from a marker image into the reference image. We show that new openings and closings are obtained, enabling intermediate results between the traditional opening (closing) and the opening (closing) by reconstruction. Some applications are studied to validate these transformations.

We describe a novel medical image coder that adopts a hybrid wavelet deringing algorithm. The proposed algorithm uses spectrum reorganization, classification, block similarity, and a deringing filter to encode medical images. The coefficients of each wavelet tree are organized into a wavelet block to reduce image complexity. Following the spatial location of the block-gathering operation, classification and block similarity are used to further reduce the total bit rate. Meanwhile, quad-tree decomposition and a set of morphological filters for reducing the ringing artifacts is presented. This set of filters employs 8 predefined morphological operations, namely, 4 structuring elements (SEs), each of which includes both a dilation and an erosion operation. The voting strategy is used to select the most suitable morphological filter for each block. Experimental results demonstrate that the proposed technique enhances the quality of the reconstructed image in both the peak SNR (PSNR) and perceptual results compared to JPEG2000 and set partitioning in hierarchical trees (SPIHT) given the same bit rate.

We present a new technique to improve the video compression ratio. In this technique, macroblock data are reordered in such a way that one block includes the important data of a macroblock, while the other three data blocks hold difference values in the horizontal, vertical, and diagonal directions. This results in reduced bit stream size because of low-valued data in the three blocks, giving a higher compression ratio. The proposed method can be easily used for error resilience applications as well. In that case, the important data block in a macroblock is transmitted on a secure channel while the remaining three blocks with difference data are sent via a lossy channel. In the case of an error in the lossy channel, the picture can still be reconstructed with a reasonably good quality using the block that contains important data transmitted on the secure channel. The proposed method generates better reconstruction quality when used at low bit rates.

Large-scale continuous media (CM) system implementations require scalable servers most likely built from clusters of storage nodes. Across such nodes, random data placement is an attractive alternative to the traditional round-robin striping. One benefit of random placement is that additional nodes can be added with low data-redistribution overhead such that the system remains load balanced. One of the challenges in this environment is the implementation of a retransmission-based error control (RBEC) technique. Because data is randomly placed, a client may not know which server node to ask for a lost packet retransmission. We design and implement three RBEC techniques that utilize the benefits of random data placement in a cluster server environment while enabling a client to efficiently identify the correct server node for lost packet requests. We implement and evaluate our techniques with a one-, two-, four-, and eight-way server cluster and across local and wide-area networks. Our results show the feasibility and effectiveness of our approaches in a real-world environment and also identify one solution as generally superior to the other two.

We present the implementation of a lossless hyperspectral image compression method for novel parallel environments. The method is an interband version of a linear prediction approach for hyperspectral images. The interband linear prediction method consists of two stages: predictive decorrelation that produces residuals and the entropy coding of the residuals. The compression part is embarrassingly parallel, while the decompression part uses pipelining to parallelize the method. The results and comparisons with other methods are discussed. The speedup of the thread version is almost linear with respect to the number of processors.

Robust hash functions are central to the security of multimedia content authentication systems. Such functions are sensitive to a key but are robust to many allowed signal processing operations on the underlying content. The robustness of the hash function to changes in the original content implies the existence of a cluster in the feature space around the original contents feature vector, any point within which getting hashed to the same output. The shape and size of the cluster determines the trade-off between the robustness offered and the security of the authentication system based on the robust hash function. The clustering itself is based on a secret key and hence unknown to the attacker. However, we show that the specific clustering arrived at by the robust visual hash function (VHF) may be possible to learn. Given just an input and its hash bits, we show how to construct a statistical model of the hash function, without any knowledge of the secret key used to compute the hash. We also show how to use this model to engineer arbitrary and malicious collisions. Finally, we propose one possible modification to VHF so that constructing a model that mimics its behavior becomes difficult.

We propose an improved autocorrelation function (ACF)-based image watermarking that is robust to combined geometric and removal attacks. ACF-based watermarking is thought of as one of the most effective watermarking schemes that resist geometric attacks. In this watermarking scheme, the autocorrelation peaks of the watermark play an important role for geometric attack estimation. The peaks, however, are vulnerable to attacks. The proposed scheme enhances the performance of ACF-based watermarking by improving the strength of the peaks. The information of an original image is used at the embedding time, so that the detector can extract strong autocorrelation peaks. Experimental results show that the proposed scheme yields better robustness than conventional ACF based watermarking against combined geometric-removal attacks.

We present a blind watermark embedding/detection algorithm to embed watermarks into H.264's I pictures. The embedded watermark can survive H.264's compression attacks with more than a 40:1 compression ratio in I pictures. One pair of predicted discrete cosine transform (DCT) coefficients within the blocks of size 4×4 are used to embed 1-bit watermark information. The embedding locations of DCT coefficients are switched from lower subbands to higher subbands in a predefined order, while the distortions between predicted values and original values are larger than a tolerable bound. With these methods, the embedded watermarks can survive H.264's compression attacks with good invisibility.

During data hiding, distortions are introduced in an original image because of quantization errors, bit replacement, or truncation at the gray-scale limit. These distortions are irreversible and visible, which is unacceptable in some applications such as medical imaging. However, the reversible watermarking technique overcomes this problem by retrieving the original image from the watermarked image. We present a novel reversible watermarking algorithm with a high embedding capacity considering the human visual system (HVS). We use the arithmetic coding technique to compress a part of the original image and store the compressed data together with necessary authentication information as the payload. The payload is then embedded within the original image with consideration of the HVS. Due to this, the watermarked image contains no perceptible artifacts. During the extraction phase, we extract the payload, restore the exact copy of the original image, and verify the authenticity. Experimental results show that our method provides a higher embedding capacity compared to the other algorithms proposed in the literature.

A scanning magnetoresistive (MR) microscope is developed for high-resolution imaging of magnetic tapes and digital media. Second-harmonic detection is used to remove thermal anomalies. We are able to image sufficient lengths of tape for authentication purposes and for data recovery from damaged samples. The second-harmonic technique provides high-contrast magnetic images of the magnetic storage media using high-resolution MR sensors from commercial hard disk drives. These images can be directly converted into the originally recorded analog audio waveforms or digital data.

Image information content is known to be a complicated and a controversial problem. We posit a new image information content definition. Following the theory of Solomonoff-Kolmogorov-Chaitin's complexity, we define image information content as a set of descriptions of image data structures. Three levels of such description can be generally distinguished: (1) the global level, where the coarse structure of the entire scene is initially outlined; (2) the intermediate level, where structures of separate, nonoverlapping image regions usually associated with individual scene objects are delineated; and (3) the low-level description, where local image structures observed in a limited and restricted field of view are resolved. A technique for creating such image information content descriptors is developed. Its algorithm is presented and elucidated with some examples, which demonstrate the effectiveness of the proposed approach.

The clustering process can be quite slow when there is a large data set to be clustered. We investigate four efficient fuzzy c-means clustering methods qFCMs, based on the quad-tree application to multispectral image feature compression and/or an aggregation process to reduce the number of exemplars for image analysis. An image is first partitioned into multiresolution blocks with variable size to extract the representative ones by homogeneity criteria. The blocks can be represented by a mean or fuzzy number to represent the image information. The first algorithm qFCMb is performed by applying only the representative blocks to a weighted FCM, which can speed up the clustering. To further improve the clustering efficiency, the reduction is done by aggregating similar examples and using a weighted exemplar in the clustering process (qFCMba). Based on the same processes used in qFCMb and qFCMba, nonhomogeneous regions including pixel information can also be supplemented to refine the clustering results, which are termed qFCMp and qFCMpa, respectively. Because of the merit of higher efficiency with the aggregation process, we recommend qFCMba and qFCMpa. A set of 14 images is used for experiments, comparison, and discussion. Performances are reported by the mean reduction rate, speedup, mean correspondence rate, and root mean square error. Results show that the mean reduction rate of both qFCMba and qFCMpa can be as high as 98% reduction in sample size. Average speedups of as much as 40 to 150 times (100 to 200 times) a traditional implementation FCM are obtained using qFCMpa (qFCMba), while producing partitions that are equivalent to those produced by FCM. On the measure of root mean square error, qFCMba is the better choice, as indicated in the experiment of clustering a noisy image.

Recently there has been a great need for efficient classification techniques in the field of medical imaging to accurately detect various human brain diseases. Extracting essential features from the magnetic resonance (MR) images of the brain is imperative for the proper diagnosis of the disease. We show that the classification success percentage is higher using features obtained from the wavelet transforms than using features obtained from the independent component analysis (ICA) for the MR human brain image data. Wavelet features that represent diseased portions are well localized and distinct, resulting in high classification accuracy. Due to their better generalization performance than neural network-based classification techniques, support vector machines (SVMs), are used for the purpose of classification. We concentrate on the stagewise classification of coronal versus sagittal images for normal versus diseased brain images, and our technique can be extended to multicategory classification, which involves various sections and disorders.

Because visual data require a large amount of memory and computing power for storage and processing, it is greatly desired to efficiently index and retrieve the visual information from image database systems. We propose efficient indexing and searching strategies for feature-based image database systems, in which uncompressed and compressed domain image features are employed. Each query or stored image is represented by a set of features extracted from the image. The weighted square sum error distance is employed to evaluate the ranks of retrieved images. Many fast clustering and searching techniques exist for the square sum error distance used in vector quantization (VQ), in which different features have identical weighting coefficients. In practice, different features may have different dynamic ranges and different importances, i.e., different features may have different weighting coefficients. We derive a set of inequalities based on the weighted square sum error distance and employ it to speed up the indexing (clustering) and searching procedures for feature-based image database systems. Good simulation results show the feasibility of the proposed approaches.

Color dropout involves converting important information in digitized color forms to black and white documents without maintaining the form background. This process is typically achieved by automatically removing all of the form lines and labels, and keeping only the information that is entered when the form is filled out. This paper expands the scope of color dropout, so that partial information from the original form is maintained for the benefit of human viewers, in a manner that does not interfere with further processing. The proposed methods assume no prior knowledge of the color form to be processed. An existing bilevel color dropout method is extended and three different multilevel color dropout approaches are developed. Results of these novel methods for processing several standard forms are presented and discussed in terms of two visual metrics—character representation and form dropout. The computational complexity of the algorithms is less than that of other common dropout methods, which makes them suitable for implementation in parallel hardware environments. A media processor implementation is presented, which outperforms general processor implementations by a factor of 11.8 on average.

We propose to set a 3-D search volume for tracking a 3-D palm motion efficiently using two cameras. If we perform template matching for right and left images independently, two points in two images do not always correspond to each other. Then, we cannot always track the correct 3-D position. Instead of finding the corresponding point in each image, we set the search volume in the 3-D space, not in the 2-D image planes, so that only valid 2-D pairs are considered in the proposed search process. The tracking process is as follows. First, we set the search volume. The 3-D coordinates of the search volume are projected on two in each image plane. We perform template matching at the projected pixel in each image. The similarity of the 3-D position is computed from two dissimilarities in the two images. We search for the position that has the maximum similarity in the search volume, and we obtain the correct correspondence result. We incorporate this technique into our tracking system, and we compare the proposed method with a method that tracks a palm motion without epipolar constraint. Our experimental results show that use of the proposed 3-D search volume makes the method accurate and efficient for tracking the 3-D motion.