Image Complexity Research Based On Geometric Features

With the rapid development of image information processing technology,image complexity has great significance in image classification and intelligent recognition technology.Because the geometrical features of the image can clearly reflect the structural features and content characteristics of the image,and the geometric features can avoid the image texture susceptible to light,noise and other interference factors.Therefore this thesis analyzes the image complexity from the perspective of image features.In this thesis,the research ongeometric features ofimage complexity is based on the image complexity evaluation method of both linear features and edge features.Based on the linear feature of the image complexity evaluation method: LSD method to detect the image of the straight line features,through expert advice on the detection of the line length of the interval division,the length of the different lengths of the number of segments as the image evaluation index,the evaluation index of normalized image is taken as the evaluation parameter of image complexity.The weight coefficients of evaluation parameters are obtained by BP neural network training method.The model of calculating the image complexity value of straight line feature is established.Then,according to the subjective evaluation method and prove the correctness of the method.Image edge evaluation method based on edge feature: Canny operator is used to detect edge of image,and the parameters of evaluation index are selected synthetically.The edge ratio,the number of connected domain and the number of line segments are taken as evaluation indexes.Parameters,normalized the parameters of the evaluation index,obtained the weight coefficient of the evaluation index parameter by weight training,and established the analysis model of the image complexity value about the edge feature.Finally the method is validated by subjective evaluation method and image complexity evaluation method based on edge fuzzy algorithm.The results show that the computational complexity of the image is consistent with human visual judgment,and the complexity of the image can be judged scientifically.