Research On Image Texture Feature Extraction Based On Complex Network

Texture is a visual feature that reflects the homogenous phenomenon in the image and describes the basic property of the slow or periodic change of the structure arrangement on the surface of the object.It exists widely in nature but is difficult to describe.It is one of the main sources of visual perception for people.Image texture feature extraction is an important research subject in the field of computer vision and pattern recognition.A large number of complex systems existing in nature can be described by different forms of complex networks.The relationship between pixels in images belongs to a kind of complex system that can be transformed into complex networks.This thesis studies the texture feature extraction in image processing and analysis from the perspective of complex network,and mainly completes the following research work:(1)An image texture feature extraction method based on complex network theory is proposed.In the modeling process of complex network,pixel points are taken as network nodes,and parameters such as radius,threshold and edge weight are set to control the network size,so as to obtain the entire evolution process of weighted complex network from sparse to dense.After the network modeling is completed,statistical features are extracted from the generated complex network set.The parameters and statistical features involved in the experiment are selected to extract measures that can effectively reflect the differences of complex network topologies corresponding to different kinds of images.(2)The complex network texture feature extraction method is integrated with the traditional texture feature extraction method.The complex network model construction of images are combined with multiple low-frequency sub-band images obtained by wavelet pyramid decomposition or the feature map of rotation invariant local binary pattern.The experimental results show that the performance of the wavelet complex network features on the Brodatz and ORL datasets are higher than that of the local binary pattern and the gray level co-occurrence matrix,and the fusion of the wavelet complex network features with the local binary pattern features on the Brodatz dataset achieves a higher classification accuracy.(3)The complex network texture feature extraction method is integrated with the HSV color space model.Based on the decomposition of H,S and V color channels,the complex network model is established for each color channel of the color image,and the network threshold evolution is carried out to extract the statistical features of the complex network set.The experimental results show that for KTH-TIPS-2B color image sets,the classification accuracy of the texture feature extraction method based on HSV color space and complex network fusion is higher than that of the feature method based on gray information,and further fusion with other features achieves better performance.