Color Image Segmentation Based On Color And Texture Feature Clustering

The Image segmentation technology has important applications in target tracking,medical detection,face recognition,fingerprint recognition,and pest identification.However,because color images contain complex color and texture information,image feature extraction and recognition are difficult,which has become a bottleneck in the development of current color image segmentation technology.Under the efforts of research scholars,color image segmentation has achieved certain results,but there is currently no universal segmentation method suitable for all color images.Aiming at this situation,this paper proposes a color image segmentation method based on color and texture feature clustering,which combines pixels with similar positions and similar color and texture features into super pixels,and clusters the super pixels.The segmentation of color images and simulation on MATLAB platform verify the effectiveness of the algorithm.The main research contents of this paper are:(1)In-depth study of color space and clustering methods.The RGB color space,HSI color space and LAB color space of color image are described.The color information of the image is characterized by the LAB color space which is the most suitable for human visual perception.The current mainstream color image segmentation methods are studied,including The traditional segmentation method,SVM,k-means,FCM and mean shifting method are used to compare and analyze various algorithms.The k-means algorithm with fast convergence speed is chosen as the theoretical basis for the clustering method of this innovative experiment.(2)An algorithm for segmenting color images combining color information and texture information is proposed.The algorithm firstly uses the Gabor filter with good response to texture in multi-scale and multi-direction to extract texture features,and uses Principal Component Analysis(PCA)to reduce the dimension.Secondly,the image is divided into large and average grids(500 by the number of grids in the experiment),and the minimum value of the texture energy gradient in the grid is selected as the initial seed point,and the LAB color space and pixel coordinates are composed of five dimensions.Vector,calculate the distance between the initial seed point and the five-dimensional vector of all the pixels in the grid,and use the distance as a similarity measure to locally cluster the pixels in the grid to form a super-pixel area block,and obtain the super-pixel area block.The feature vector of the cluster center.Finally,the k-means algorithm is used to re-cluster the super-pixel region according to the Euclidean distance of the feature vector,and the segmented target region is obtained.(3)The proposed algorithm and the current mainstream color image segmentation method are compared with the images in the MIT color texture library and the Segbench natural landscape image library.It is proved that the target region is accurate under the premise that the running time is close to other algorithms,less noise,from a subjective perspective,the segmentation results are smoother.This method of clustering color images by using color information and texture information improves the shortcomings of the current mainstream segmentation methods to a certain extent.Through the research of this subject,the key techniques of feature extraction and clustering required to segment color images are mastered,which lays a theoretical foundation for future academic research and engineering practice.