Image Edge Feature Extraction Algorithms And Applications

Edge is important visual information. Image edge extraction plays an important role in image processing and machine vision. In this paper, image edge feature extraction algorithms are thoroughly studied, and many experimentations and applications are provided to validate the effectively of the presented algorithms. The main research work and innovative results are as follows:1. Spline wavelet based adaptive thresholds multi-scale edge extraction algorithm.Image edge extraction occupies an important role in Marr's computer vision system. Because of its complexity and technical restrictions, digital image edge extraction has not been a perfect solution. The wavelet transform has a good time-frequency localization characteristics, it is particularly suited to the analysis and processing of unbalanced signals. Since Mallat put forward fast algorithm of wavelet transformation in 1980's, wavelet transform is widely used in many image processing areas, and have achieved good results. There has been some wavelet transform based image edge extraction methods, but it is very difficult for them to satisfy Canny criteria and asymptotically optimal at the same time. In addition, most of these methods use only one threshold value, and can't obtain good effects for images with dissimilarity edges.Based on the characteristics of wavelet transform, a B-spline wavelet based adaptive thresholds multi-scale edge extraction algorithm is studied. Firstly, the cubic B-spline smoothing filter operator is designed to perform multi-scale filtering. The obtained multi-scale wavelet transform is combined with adaptive thresholds to extract image edge at different scales. Then using the multi-scale characteristic of the edge information and synthesizing multi-scale edge, the final image edge with only a pixel width is obtained. The method is confirmed by experimental results which demonstrate that the algorithm can effectively restrain noise, but also to detect the edge of rich details.2. Embedded confidence combined multi-scale edge extraction algorithm. Classical edge extraction methods are mainly based on the selective criteria of grads vector amplitude, but the information offered by amplitude have fixed uncertainty for the effect of data pattern and edges size. If the information which is not included in grads vector amplitude calculation is used to estimate the comparability of data pattern and ideal edges model, the uncertainty will be greatly reduced.Based on the above analysis, the paper proposes an edge extraction algorithm with embedded confidence. Firstly, the disadvantages of traditional continuous criteria of Canny are indicated, based on which discrete criteria of Canny and the optimal filter of the criteria are discussed. According to discrete criteria of edge extraction, numerical method is used to get the optimal filter and smooth operator. Secondly the smoothing filter operator is designed and combines with the embedded confidence to detect edges. Thirdly, with full use of the multi-scale character of the edge information, which has rich details and more accuracy with small scales and higher Signal-to-noise-ration with large scales, the detected multi-scale edges are synthesized to obtain accurate single pixel wide edges. The experimental results demonstrate that the algorithm has strong anti-jamming capability and better noise suppression ability to effectively extract the real edge from jamming.3. Multi-scale morphological edge extraction algorithm.Classical edge extraction methods are always hypersensitive to noise and have low ability to anti-jamming as they introduced all kinds of differential operators; such as Sobel operator, Prewitt operator etc. Neoteric edge extraction operators need plenty of numeration to smooth original image and easy to blurred image edges, such as Log operator, Canny operator etc. Based on the above issues, a multi-scale adaptive weighted morphological edge extraction algorithm is studied. The paper firstly discusses some applications of mathematical morphology in edge extraction, and then a new multi-scale adaptive weighted morphological edge extraction algorithm is presented to detect image edge. Secondly the scale adjustment of the morphological structural elements is performed. The more ideal image edges under the noise disturbance are obtained by integrating the edge characteristics of various scales. The experimental results demonstrate that the presented algorithm can avoid complicated calculations and blurred edges by smooth operation of some edge extraction operators. It also has good real-time performance and better noise suppression ability to extract clear edge.4. Wavelet edge extraction algorithm with signal registration technology.Edges are the mark of region boundaries. In practical applications, it is more important to detect continuous edges than detect discrete points. Vehicle feature extraction is a key point of vehicle type recognition in Intelligent Transportation System (ITS), which needs continuous and smooth edge extraction results with noise be effectively restrained to describe vehicle feature. Stereo matching is an important part of robot vision, which need accurate and continuous results of interesting target with disturbance be effectively restrained to improve 3D reconstruction effect.To meet the above two requirements of vehicle type recognition and stereo matching, a signal registration technology combined wavelet multi-scale edge extraction algorithm is proposed. The method firstly design wavelet multi-scale edge extraction algorithm to detect image edges based on the characteristic of wavelet transform. Then the discontinuous edges are connected into a contour line with signal registration technology to weaken the influence of noise or disturbance. The experimental results demonstrate that the method can not only precisely detect clear and continuous image edges, but also effectively suppress noise and disturbance. It can be applied to many fields such as vehicle recognition and tracking, robot vision, and so on. 5. Application of multi-scale image edge extraction technology in image compression.Image compression is an important technology in image transmission and storage. Under low bit rate, images reconstructed form wavelet-based compression algorithms have blurred edge, which is called Gibbs phenomenon. This phenomenon hampers the recognition of objects in the image.To overcome Gibbs phenomenon in wavelet-based image compression, the paper presents an edge preserving embedded image compression algorithm to improve the visual appearance and reconcilability of compressed images at low bit rates. Edge informations are enhanced before compression coding by certain weighting factors. Experimental and 3D reconstruction results demonstrate that the algorithm can yield high performance, and Gibbs phenomenon is effectively overcome.