Ant Colony Image Edge Detection Based On Multi-window And Parallel Algorithm

Edge detection is a heated research direction in image processing. Since the research of complex images, such as satellite image and remote sensing image have been conducted more and more deeply, the amount of data need to be processed also increases rapidly. To extract the important information in these images as fast as possible, there is the urgent demand of real-timing in processing these kinds of images. In this thesis, we study some edge detection algorithms using multi-window strategy and analyze the related parallel algorithms. The original image is divided into many independent sub-images, which could be assigned to different processors applying multi-window mechanism to achieve parallel implementation. At the same time, in every window that divided from the original image, we use the ant colony algorithm to detect the edge due to its potential parallelism.To provide a parallel algorithm for detecting edges of image, we firstly propose a multi-state ant colony edge detection algorithm based on same size window division. According to the locality of the image, this algorithm divides the original image into multiple windows and sets different thresholds and parameters in different windows to initialize independent ant colony algorithms for detecting image edges. As a result, every window could apply individual detecting algorithm to achieve parallel running and every window is allocated different parameters to improve the efficiency and effectiveness of detection. This algorithm also introduces the multi-state ant colony to improve the search efficiency. Specifically, it will set the scout for a primary detection to remove the non-edge regions in the image. The simulation experiment results show that this optimized algorithm is more efficient than the traditional algorithms.To enhance the poor flexibility and adaptability of the same size window division mechanism, we introduce an improved multi-window edge detection algorithm, the division strategy of which is based on the information entropy. The information entropy is used to reflect the amount of information contains in an image. This improved algorithm could automatically divide the original image into multiple windows with different numbers and sizes according to the information entropy of this image. It also applies adaptive second division mechanism, which could be used to divide the image into windows with similar amount of information. Then the edge detection algorithm is run individually in every window. We can see that this algorithm could improve the efficiency of edge detection and pave the way for the parallel detection algorithm. According to our simulation experiments, this new algorithm is faster and more efficient than the previous algorithm.Traditional detection algorithm has encountered the bottle neck in the efficiency in detecting complex images to satisfy the real-timing requirement. Therefore, we present a parallel detection algorithm based on multi-window strategy, which modifies the traditional sequential algorithm into parallel algorithm. It applies the double parallel model based on the GPU parallelism, which divides the image into multiple windows in the coarse-grain way and the parallelism of ant colony algorithm in different windows in the fine-grain way. This double parallel model could greatly improve the process speed in detecting complex images to handle more images in the same time and space cost. The simulation experiment proves that the parallel algorithm is more efficient than the sequential algorithm while the detection quality is also guaranteed in the parallel algorithm setting.