No-reference image quality evaluation method based on multi-branch similarity network

Abstract

The invention discloses a no-reference image quality evaluation method based on a multi-branch similarity network. The no-reference image quality evaluation method comprises the steps: constructing amulti-branch convolution module and a similarity fusion module, constructing a multi-branch similarity network, generating a no-reference training data set and a no-reference test data set, training amulti-branch convolution feature similarity network, and outputting a quality evaluation score value of each distorted image in the test set. According to the invention, the multi-branch convolutionmodule adaptively extracts the hierarchical features of the image, reduces the correlation between the image features and the image content through similarity fusion, is more accurate in result when evaluating the quality of a reference-free image, and is wider in application range.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to View More

AI Technical Summary

Benefits of technology

This patented technology allows for better understanding by create an algorithm called Multiplex Branch Convolution Module (MCM) that automatically adjusts how well images look on them based upon their characteristics such as color or texture. It also uses similarity analysis techniques to find correlations between these attributes and evaluate the overall effectiveness of this process. Overall, this innovation improves accuracy while reducing redundancy issues associated with previous methods like manually selecting representative pixels during testing.

Problems solved by technology

Technological Problem addressed in this patents relating to non-representational image qualification involves overcoming limitations associated with current techniques like manual or automatic focus adjustments due to lack of sufficient data sources. Existing approaches involve analyzing entirety pixels without any specific areas containing relevant details about each pixel. These technical drawbacks include inconvenience caused by unrelated background colors, difficulty in selecting representative regions during analysis, potential overlap among different spectral components, and difficulties in accurately representing complex objects within the image.

Images