Retrieval of texture feature information data for small differentiated images based on adaptive feedback mechanism
To solve the problem of poor texture feature mining performance caused by texture similarity and noise in small differentiated images,the small differentiated image texture feature mining method was designed with combining adaptive feedback and local binary mechanism.The standard cut strategy was used to model each point in the image data as node,and the similarity between the two points was calculated by the weights of connecting lines among the nodes.The support vector machine was adopted to train image attribute parameters for classifying image attributes and image categories.Using the skip connection method to transmit image data,the data was introduced into convolutional neural network to remove image noise.Taking the pixel value of the center point as feedback factor,the adaptive feedback judgment conditions were created,and the local binary patterns were used to achieve small differential image texture feature mining.The experiments were conducted on the MATLAB platform to analyze the convergence of convolutional neural networks,the number of spectral texture units in images,the average accuracy and the image data matching.The analysis results show that as the number of iterations is increased,the accuracy loss is gradually decreased and converged to stable value,which achieves expected training effect.The proposed method can mine over 3 800 spectral texture units in images,which are more in line with human visual information.The average accuracy is 0.87,and the average values of accuracy@1,accuracy@5 and accuracy@10 are 0.90,0.84 and 0.85,respectively.The mining time is less than 5 seconds,and the image data matching degree is higher than 90.3％,which verifies that the proposed method can play important role in image texture feature recognition operations.
万方数据
维普
