Image fuzzy model parameter analysis method based on depth learning

Abstract

The invention designs a fuzzy model parameter analysis method based on depth learning. The method is characterized by including: step 1, adopting a depth belief network structure for extracting and classifying fuzzy features, namely using a semi-supervised depth belief network to project an input image block to a difference feature space, and then classifying each feature; step 2, recognizing parameters of a fuzzy kernel, namely extracting from the edge of a transform domain to guarantee that the depth belief network structure confirms fuzzy parameters with high accuracy. By the image fuzzy model parameter analysis method based on depth learning, the fact that the depth belief network structure confirms the fuzzy parameters with high accuracy can be guaranteed; according to repeated trial verification in multiple image databases such as Berkeley segmentation, Pascal VOC 2007 and the like, favorable performances superior to those of the existing best fuzzy estimation method are achieved.

Description

Technical field [0001] The invention belongs to the technical field of image blur analysis, and particularly relates to an image blur model parameter analysis method based on deep learning. Background technique [0002] Blurred image repair is a process of reconstructing the original high-quality image using insufficient information of the existing degradation model. Image deblurring methods can be divided into blind deblurring and non-blind deblurring. The non-blind defuzzification method requires prior knowledge of the fuzzy kernel and its parameters, while in the blind defuzzification method, it can be assumed that the fuzzy operator is unknown. In various situations in practical applications, the Point Spread Function (PSF) is unknown, so the application range of non-blind deblurring is much narrower than that of blind deblurring. Blind image deblurring algorithms can be divided into two categories: multi-image and single-image deblurring methods. In practical applications...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for analyzing parameters of the image blur model based on deep learning. The present invention aims at the problems of different blur types and blur parameter sizes in the same existing image. First, Use the two-step deep belief network structure to first classify the fuzzy type, and then identify the parameters of the fuzzy kernel to ensure that the deep belief network structure confirms the fuzzy parameters with high accuracy

Images