Research on Intelligent Algorithm for Image Quality Evaluation Based on Image Distortion Type and Convolutional Neural Network

Abstract

In order to detect different images under different distortion conditions, this paper classifies image distortion types and proposes video image distortion detection and classification based on convolutional neural networks. By segmenting the input video image to obtain a small block image, and then using the active learning feature of the convolutional neural network, the positive and negative cases are equalized, the adaptive learning rate is slowed down and the local minimum problem is solved. The type of distortion is mainly predicted by the SoftMax classifier image, and then the video image prediction type obtained by the majority voting rule is used. The objective quality evaluation algorithm based on image distortion type and convolutional neural network is analyzed. Using LIVE (simulation standard image library) and actual monitoring video library, the final accurate results of the two performance tests are not much different. The overall classification accuracy rate is significantly higher than other algorithms. After the positive and negative case equalization and adaptive learning rate are introduced, it is found that the CNN classification accuracy can be significantly improved. The final test results also confirm that this method can actively learn image quality features, and improve the accuracy of video image classification detection. It is applicable to image quality detection of all video image distortion conditions, and has strong practicability and robustness.