Bidimensional convolution is a low-level processing algorithm of interest in many areas, but its high computational cost constrains the size of the kernels, especially in real-time embedded systems. This paper presents a hardware architecture for the ASIC-based implementation of 2-D convolution with medium–large kernels. Aiming to improve the efficiency of storage resources on-chip, reducing off-chip bandwidth of these two issues, proposed construction of a data cache reuse. Multi-block SPRAM to cross cached images and the on-chip ping-pong operation takes full advantage of the data convolution calculation reuse, design a new ASIC data scheduling scheme and overall architecture. Experimental results show that the structure can achieve 40× 32 size of template real-time convolution operations, and improve the utilization of on-chip memory bandwidth and on-chip memory resources, the experimental results show that the structure satisfies the conditions to maximize data throughput output , reducing the need for off-chip memory bandwidth.
This paper proposes a low-cost FPGA architecture of Speed-Up Robust Features (SURF) algorithm based on OpenSURF. It optimizes the computing architecture for the steps of feature detection and feature description involved in SURF to reduce the resource utilization and improve processing speed. As a result, this architecture can detect feature and extract descriptor from video streams of 800x600 resolutions at 60 frames per second (60fps). Extensive experiments have demonstrated its efficiency and effectiveness.
The infrared small target’s detection and tracking are important parts of the automatic target recognition. When the camera platform equipped with an infrared camera moves, the small target’s position change in the imaging plane is affected by the composite motion of the small target and the camera platform. Traditional detection and tracking algorithms may lose the small target and make the follow-up detection and tracking fail because of not considering the camera platform’s movement. Moreover, when there exist small targets with different motion features in the camera’s view, some detection and tracking algorithms can’t recognize different targets based on their motion features because there are no trajectories in a unified coordinate system, which may lead to the true small targets undetected or detected incorrectly . To solve those problems, we present a method under the condition of moving camera platform. Firstly, get the camera platform’s motion information from the inertial measurement values, and then decouple to remove the motion of the camera platform itself by means of coordinate transformation. Next, estimate the trajectories of the small targets with different motion features based on their position changes in the same imaging plane coordinate system. Finally, recognize different small targets preliminarily based on their different trajectories. Experimental results show that this method can improve the small target’s detection probability. Furthermore, when the camera platform fails to track the small target, it’s possible to predict the position of the small target in the next frame based on the fitted motion equation and realize sustained and stable tracking.
Proc. SPIE. 9675, AOPC 2015: Image Processing and Analysis
KEYWORDS: Digital signal processing, Logic, Clocks, Detection and tracking algorithms, Imaging systems, Image processing, Image resolution, Field programmable gate arrays, Control systems, Image transmission
In real-time image processing, with the improvement of resolution and frame rate of camera imaging, not only the requirement of processing capacity is improving, but also the requirement of the optimization of process is improving. With regards to the FPGA + DSP architecture image processing system, there are three common methods to overcome the challenge above. The first is using higher performance DSP. For example, DSP with higher core frequency or with more cores can be used. The second is optimizing the processing method, make the algorithm to accomplish the same processing results but spend less time. Last but not least, pre-processing in the FPGA can make the image processing more efficient. A method of multi-resolution pre-processing by FPGA based on FPGA + DSP architecture is proposed here. It takes advantage of built-in first in first out (FIFO) and external synchronous dynamic random access memory (SDRAM) to buffer the images which come from image detector, and provides down-sampled images or cut-down images for DSP flexibly and efficiently according to the request parameters sent by DSP. DSP can thus get the degraded image instead of the whole image to process, shortening the processing time and transmission time greatly. The method results in alleviating the burden of image processing of DSP and also solving the problem of single method of image resolution reduction cannot meet the requirements of image processing task of DSP.
Image matching is a fundamental task in computer vision. It is used to establish correspondence between two images
taken at different viewpoint or different time from the same scene. However, its large computational complexity has
been a challenge to most embedded systems. This paper proposes a single FPGA-based image matching system, which
consists of SIFT feature detection, BRIEF descriptor extraction and BRIEF matching. It optimizes the FPGA architecture
for the SIFT feature detection to reduce the FPGA resources utilization. Moreover, we implement BRIEF description and
matching on FPGA also. The proposed system can implement image matching at 30fps (frame per second) for 1280x720
images. Its processing speed can meet the demand of most real-life computer vision applications.
Blind motion deblurring from a single image is a challenging ill-posed problem. Significant progress has been made
since blur kernel estimation method using salient edge prediction on transfer region is proposed. However, as selection
rule for points to estimate the blur kernel has not been researched deeply, some texture and noise points were taken into
account for blur kernel estimating, which makes the existing methods not robust enough. This paper propose a robust
motion deblurring algorithm using salient edge prediction on transfer region, which employs a new metric to select
transfer region points for kernel estimation. A novel kernel refinement method with hysteresis thresholding is also
proposed and adopted by the algorithm to reduce the kernel noise. Extensive experiments show that the algorithm
achieves good results, while both the new metric and the novel kernel refinement method improve robustness of the
By analyzing the characteristics of infrared focal plane array image, an improved implementation of infrared focal
plane image enhancement algorithm based on FPGA is proposed, with limited FPGA memory resources for gray-scale
stretching. Experiment results show that the implementation is easy on FPGA with low FPGA memory without extra
memory devices. Moreover, it is flexible and effective for improving gray contrast of the interested region of the
image, and proved to meet the requirements of infrared focal plane detector for image enhancement showing great
This paper presents a novel fast template matching algorithm based on context prediction. The predicted regions are
those windows that contain the current entire sub-window. Comparison skipping or comparison terminating is executed
when a low bound of distance which has been calculated between the template and the window exceeds the threshold.
Experimental results and theory analyses prove the proposed method is faster than the conventional fast template
matching method, strictly guaranteeing the same accuracy and up to maximal twenty times faster than the SSDA.
Two-dimensional entropic thresholding methods apply gray-level spatial correlation to thresholding and achieve much better performance than 1-D methods, while suffering from large time consumption. To utilize gray-level spatial correlation in thresholding with less time consumption, we define and describe a new entropic thresholding approach employing the gray-level spatial correlation (GLSC) histogram. The GLSC histogram is determined using the gray value of the pixels and the number of their neighboring pixels of similar gray value, which is different from a 2-D histogram. During the entropic criterion function computation, the entropy yielded by different elements in the GLSC histogram is weighted by a nonlinear weighting function, which we suggest. In the experiment, Kapur's 1-D method and three 2-D methods reported by Abutaleb and Sahoo are employed for comparision. Experiments on many real-world images demonstrate that the proposed method yields equivalent or even better results than 2-D ones while saving time remarkably and significantly outperforms Kapur's 1-D method without too much more time consumption generally.
This paper presents a new method for multi-focus image fusion. In the method, the source images are first decomposed
into blocks, and the decomposed images are then combined by the use of adaptive Wiener filter. Effects of the block size
and threshold are analyzed, and comparison with wavelet transform based method is done. Experimental results show
that the proposed method is comparative to wavelet transform based methods for the images without noise, while this
method is computationally simpler, and can be implemented in real-time applications. Experimental results also show
that under noise circumstances, additive noise or multiplicative noise, the proposed method is obviously superior to the
wavelet based method.
Proc. SPIE. 6786, MIPPR 2007: Automatic Target Recognition and Image Analysis; and Multispectral Image Acquisition
KEYWORDS: Edge detection, Digital signal processing, Image processing, Image resolution, Data processing, Signal processing, Light sources and illumination, Image filtering, Nonlinear filtering, Filtering (signal processing)
Downward looking scene matching is an important technique of the aircraft automation guidance. To solve the
heterogeneous images scene matching problem, we present two effective methods based on intensity-based correlation in
this paper. One is to search the real match position based on the feature of the peak on the correlation surface. We
propose a criterion to search the proper matching. The other is to use a non-linear filter to pre-process the images, which
reduces the influence of ambient lighting while keeping the necessary image details since the contour of the scene is the
stable and unchanged feature. Also, we use a Fourier analysis to explain the contribution of different frequency spectrum
in the correlation. By using this frequency information, we propose a simpler kernel filter method based on pre-process,
which has the similar performance with non-linear filter pre-process but has less computation complexity. This simple
kernel is more suitable for the embedded DSP real-time application.
2-D Convolution is a simple mathematical operation which is fundamental to many common image processing operators.
Using FPGA to implement the convolver can greatly reduce the DSP's heavy burden in signal processing. But with the
limit resource the FPGA can implement a convolver with small 2-D kernel. In this paper, An FIFO type line delayer is
presented to serve as the data buffer for convolution to reduce the data fetching operation. A finite state machine is
applied to control the reuse of multipliers and adders arrays. With these two techniques, a resource limited FPGA can be
used to implement a larger kernel convolver which is commonly used in image process systems.
To solve the matching problem about the images got from different sensors, we present a new method based on intensity-based correlation. After analyzing the real match position and false match position from the correlation surface, we found a new method to search the real match position based on the feature of the peak on the correlation surface. Feature used in this method includes relative height of the peak, width of the peak and distinctness degree of the peak. Experiments show that this method is effective on the condition of proper sensed image size and resolution.
In this paper, the Fourier translation of 1-Dimension continuous signal is used to analysis the different frequency spectrum. The convolution of two signals can be expressed as the convolution of Fourier series of these two signals¡¯. After some translations and ignoring some secondary factors, the formula shows that in the correlation of two signals, the higher frequency of the images causes the narrow peak on the surface of the correlation. Comparing the original image and the correlation surface, we found the narrow peak on the correlation surface indicated the real matching position. All these shows that the stable unchanged feature usually contain in the higher frequency of the different images. To solve these matching problem of multi-spectral images, two methods are proposed, one is to do pretreatment (enhance the high frequency of the images to eliminate inconstant factors), the other is to search the narrow peak on the correlation surface. All these two methods are the same effect to locate the real matching position.
In the farther experiment, we found the matching problem not only between different spectral images, but also between the same spectral images but got at different time and different conditions had the same principle. The frequency analysis method can be extended to the problem of heterogeneous images matching.
Relaxation matching is one of the most relevant methods for image matching. The original relaxation matching technique using point patterns is sensitive to distortions such as missing or spurious points and random errors. In this paper, we present a robust point pattern relaxation matching technique based on least median of squared errors estimation to improve the original point pattern relaxation matching technique to be robust to missing or spurious points and random errors. Experimental results with large simulated images and real images demonstrate the effectiveness and feasibility of the method to perform point pattern relaxation matching with missing or spurious points and random errors.