Research On Color Image Segmentation Method Based On Sparse Representation

Image recognition has a wide range of applications in many fields,such as autopilot and medical diagnosis.The result of image segmentation will greatly affect the accuracy and correctness of image recognition.Previous researches on image segmentation mainly focused on the study of grayscale images.However,because the color image is more in line with human visual characteristics,only the study of grayscale image segmentation can not meet the actual needs.Therefore,people pay more and more attention to the study of color image segmentation.In recent years,researchers have confirmed that most of the gray image segmentation algorithms can also be used for color image segmentation,but because of the color image contains more complicated information than the gray image,the grayscale image segmentation algorithm has already unable to achieve the intended purpose,that is why studying color image segmentation is so important.Sparse representation theory has an increasingly important influence in the field of image recognition.It has extensive and in-depth applications in image denoising,image reconstruction,pattern recognition,and so on,and has achieved better results than traditional methods.In this paper,first,the research background and status quo of color image segmentation and sparse representation are briefly introduced.Then some of the color image segmentation methods which are more commonly used in recent years are reclassified.These methods are roughly divided into threshold-based,cluster-based,region-based,edge-based and based on specific theories.Some of the more classic segmentation algorithms are introduced in detail,and the advantages and disadvantages of these algorithms are analyzed.Secondly,we introduced the related theory of sparse representation systematically,and discussed the K-SVD dictionary learning algorithm in depth,which provides the theoretical basis for the following algorithms and experiments.Based on the above theoretical research,we introduced the sparse representation into color image segmentation,and proposed a new color image segmentation algorithm which is based on sparse representation.The algorithm uses the sparse representation to improve the color image edge detection algorithm and then achieves the purpose of the final segmentation.In the sparse representation-based color image edge detection framework,the image is filtered to reduce the noise firstly,and then the pre-processed image is sparse coding and dictionary learning using the fixed dictionary DCT and the dictionary learning algorithm K-SVD respectively.After learning through iterative updating,an optimal dictionary that can represent the image is obtained.After that,the image is reconstructed and detected using the obtained dictionary,and then the final edge detection image can be obtained.Based on the above framework,the color image is further segmented to obtain a better segmentation result.Experimental results show that the proposed method can effectively detect the edges of color images and segment the color images based on the detected edges.