The problem of extracting continuous structures from images is a difficult issue in early pattern recognition and image
processing. Tracking with contours in a filtering framework requires a dynamical model for prediction. Recently,
Particle filter, is widely used because its multiple hypotheses and versatility within framework. However, the good
choice of the propagation function is still its main problem. In this paper, an improved particle filter, EM-PF algorithm is
proposed which using the EM (Expectation-Maximization) algorithm to learn the dynamical models. The EM algorithm
can explicitly learn the parameters of the dynamical models from training sequences. The advantage of using the EM
algorithm in particle filter is that it is capable of improve tracking contour by having accurate model parameters. Though
the experiment results, we show how our EM-PF can be applied to produces more robust and accurate extracting.
Proc. SPIE. 6027, ICO20: Optical Information Processing
KEYWORDS: Target detection, Digital signal processing, Detection and tracking algorithms, Image processing, Field programmable gate arrays, Image analysis, Data processing, Signal processing, Target recognition, Data communications
Template matching is the process of searching the present and the location of a reference image or an object in a scene image. Template matching is a classical problem in a scene analysis: given a reference image of an object, decide whether that object exists in a scene image under analysis, and find its location if it does. The template matching process involves cross-correlating the template with the scene image and computing a measure of similarity between
them to determine the displacement. The conventional matching method used the spatial cross-correlation process which is computationally expensive. Some algorithms are proposed for this speed problem, such as pyramid algorithm, but it still can't reach the real-time for bigger model image. Moreover, the cross-correlation algorithm can't be effective when the object in the image is rotated. Therefore, the conventional algorithms can't be used for practical purpose. In
this paper, an algorithm for a rotation invariant template matching method based on different value circular projection target tracking algorithm is proposed. This algorithm projects the model image as circular and gets the radius and the sum of the same radius pixel value. The sum of the same radius pixel value is invariable for the same image and the any rotated angle image. Therefore, this algorithm has the rotation invariant property. In order to improve the matching speed and get the illumination invariance, the different value method is combined with circular projection algorithm. This method computes the different value between model image radius pixel sum and the scene image radius pixel sum so that it gets the matching result. The pyramid algorithm also is been applied in order to improve the matching speed. The high speed hardware system also is been design in order to meet the real time requirement of target tracking system. The results show that this system has the good rotate invariance and real-time property.
Hough transform is recognized as a powerful tool in shape analysis which gives good results even in the presence of noise and the disconnection of edge. However, traditional Hough transform can only detect the lines, cannot give the endpoints and length of the line segments and it is vulnerable to the quantization errors. Based on the analysis of its limitations, Hough transform has been improved in order to detect line segment feature of targets. The algorithm aims to avoid the loss of spatial information, as well as to eliminate the spurious peaks and fix on the line segments endpoints accurately, which can expediently be used for the description and classification of regular objects. The method consists of 6 steps: 1. setting up the image, parameter and line-segment spaces; 2. quantizing the parameter space; 3. applying the standard Hough transform equation to every point of the input image edge, and extracting a group of maximums according to the global threshold; 4. according to the local threshold, eliminating spurious peaks which are caused by the spreading effects; 5. fixing on the endpoints of the segments according to the dynamic clustering rule; 6. merging the segments whose extreme points are near. Experiment results show the approach not only can recognize regular geometric object but also can extract the segment feature of real targets in complex environment. So the proposed method can be used in the target detection of complicated scenes, and will improve the precision of tracking.
In many computer vision tasks, in order to improve the accuracy and robustness to the noise, wavelet analysis is preferred for the natural multi-resolution property. However, the wavelet representation suffers from the dependency of the starting point of the sampled contour. For overcoming the problem that the wavelet representation depends on the starting point of the sampled contour, the Zernike moments are introduced, and a novel Starting-Point-Independent wavelet coefficient shape matching algorithm is presented. The proposed matching algorithm firstly gains the object contours, and give the translation and scale invariant object shape representation. The object shape representation is converted to the dyadic wavelet representation by the wavelet transform. And then calculate the Zernike moments of wavelet representation in different scales. With respect to property of rotation invariant of Zernike moments, consider the Zernike moments as the feature vector to calculate the dissimilarity between the object and template image, which overcoming the problem of dependency of starting point. The experimental results have proved the proposed algorithm to be efficient, precise, and robust.
This paper discusses image's object edge extracting of threshold value. First, this paper introduces fundamental principle of gray image object edge extracting, after discussing the influence of object edge image's gray scale distributing characteristic and choice of threshold value to character extracting and target identification, we bring forward a sort of method, which using object edge fuzzy taxonomy and maximum fuzzy entropy theory to auto choose threshold value in the case that membership function is given, then we have a test of threshold value segmentation to edge gray image in different contrast, the result indicates this method has preferable scene adaptive faculty.
It is difficult to choose eigenvectors when neural network recognizes object. It is possible that the different object eigenvectors is similar or the same object eigenvectors is different under scaling, shifting, rotation if eigenvectors can not be chosen appropriately. In order to solve this problem, the image is edged, the membership function is reconstructed and a new threshold segmentation method based on fuzzy theory is proposed to get the binary image. Moment invariant of binary image is extracted and normalized. Some time moment invariant is too small to calculate effectively so logarithm of moment invariant is taken as input eigenvectors of BP network. The experimental results demonstrate that the proposed approach could recognize the object effectively, correctly and quickly.
To absolve image blur caused by camera's high speed, fixed-point DSP TMS320C6416 and wiener filtering algorithm are adopted to stabilize image. On the premise that cameras have uniform motion, restoration model for moving blurred image was researched, the principle of Wiener filtering is explained, the key technologies of moving blurred image restoration with TMS320C6416 are introduced: how to complete FFT calculation of 32bit and how to exert DMA function of DSP to enhance processing speed, ring effect and ghost effect in restored image are explained, and how the parameter λ of wiener filter affects ring effect and ghost effect is discussed, in addition to this, other reasons of ring effect and ghost effect are analyzed detailed. Experiments shows that One TMS320C6416 chip can restore seven frames per second, there is great hope to realize real-time restoration if adopting multi-DSP or the FFT completion by hardware.
Proc. SPIE. 6027, ICO20: Optical Information Processing
KEYWORDS: Digital signal processing, Digital image processing, Image processing, Video, Field programmable gate arrays, Image analysis, Digital imaging, Signal processing, Video processing, Data communications
In order to realize the auto focusing technique based on digital image processing, a video image processing system is designed by using high-performance digital signal processor TMS320VC5509 as the core, field programmable gate array FPGA for preprocessing of image. At the same time, it is proposed to use vector norm sum of image gray gradient as evaluation functions of digital image. It has the characteristics of good unbiasedness, powerful unimodality, etc., and can be applied to evaluation of defocusing. The hardware composition of the theory behind the auto focusing system, and the application of the evaluation function are discussed in details.
Proc. SPIE. 5623, Passive Components and Fiber-based Devices
KEYWORDS: Target detection, Digital signal processing, Digital image processing, Detection and tracking algorithms, Image processing, Field programmable gate arrays, Image analysis, Signal processing, Target recognition, Evolutionary algorithms
In order to resolve the contradiction between real-time and arithmetic complex in television tracking capture system, the paper discusses a real-time target track processing system which is constructed by high performance DSP chipset TMS320C6416 as core digital processor, huge reprogrammable logic chipset CPLD as system logic controller and field reprogrammable array FPGA as image preprocessing chipset to sampled video digital image. In the same time, the author also improved target capture arithmetic by introducing a kind of fast image correlation matching arithmetic based on evolutionary algorithms. Major parts put on hardware construct, working theory and new image correlation matching algorithms. Furthermore the comparison of the performance provided by this method with conventional matching algorithms is discussed. Theoretical analysis and simulation results show that the proposed algorithm is very effective.