Multi-threshod Image Segmentation Based On Chaos Theory

Image segmentation, feature extraction and target recognition are three major tasks from low levels tohigh levels in image processing. Image segmentation is the foundation for feature extraction and targetrecognition. The segmentation results have directly influence on subsequent feature extraction and targetrecognition. Multi-target recognition is widely used in image processing and target recognition because it issimple and effective, so the research on image segmentation has practical significance. The traditionalmulti-threshold image segmentation methods have some disadvantages, such as high complexity, slowcomputing speed and the number of split classes determined in advance, etc. Chaos belongs to non-lineardynamical systems, which has some good points: ergodicity, convergence and random unpredictability. Theapplication of chaos theory in multi-threshold image segmentation will help improve the performance ofmulti-threshold image segmentation and speed up the running time. This paper makes the application ofchaos in multi-threshold as the focus. The main works we have done are as follows:(1) Fisher multi-threshold image segmentation based on chaotic particle swarm is proposed. First, thepotential function is used to determine the number of image segmentation. The potential function fits thehistogram, whose advantages are simple and fast computing. Combined with the advantages, we usepotential function to determine the number of image segmentation. Second, the fisher criterion is simplifiedin order to reduce unnecessary intermediate calculation and then, it is extended to the multi-thresholdimage segmentation. Last, chaotic particle swarm optimization algorithm is used to find the best thresholds.Particle swarm optimization algorithm is simple and it has fast computing time, but it is prone to the phenomenon of ‘premature’. Therefore, particle swarm algorithm based on chaos is used. Imagesegmentation results show that the algorithm has better results in computing speed and segmentation effect.(2)Multi-threshold image segmentation algorithm based on spatio-temporal chaos is improved. Thespatio-temporal chaos algorithm utilizes the synchronization process of global coupled map correspondingto the process of data clustering to complete the image segmentation. This paper has improved thealgorithm. First, the initialization data is improved: coupled map lattice is initialized by theone-dimensional histogram vector. Using the linear programming ideological, the original gray matrix isassigned to the sum of gray matrix and probability matrix, so that we can make full use of the specificimage information. Then, relationship matrix is re-assigned after each iteration to meet the requirement ofthe binary matrix. Last, parameters used in the relationship matrix are discussed and the improvedalgorithm is tested to prove its effectiveness.(3)Spectral clustering image segmentation based on chaotic mapping is proposed. First, gray matrix isused as the input data instead of pixel matrix, which can save a lot of storage space and speed up thecomputing time. Second, k-mean clustering in the spectral feature space clustering is easy to fall into localsolution, so we use chaotic Logistic map to adjust the optimal thresholds. Finally, the parameters used inthe similarity matrix are discussed. The image segmentation results show the good segmentation results.