Face image quality evaluation method based on multidirectional evaluation standard and system thereof

Abstract

The invention provides a face image quality evaluation method based on a multidirectional evaluation standard and a system thereof. The method comprises the steps of acquiring a face image as trainingdata; classifying the training data according to degrading characteristics with different dimensions; performing preliminary evaluation and marking on each kind of images in the classified training data; establishing a branch task which corresponds with each kind of dimension characteristics according to the classification result of the degrading characteristics with different dimensions; establishing a quality evaluation model, performing training, performing multi-branch task predicating on an input to-be-evaluated face image according to the quality evaluation model after training, therebyfinishing face image quality evaluation. According to the method and the system, the image characteristics are extracted by means of a deep neural network, and furthermore comprehensive quality evaluation to multiple image quality affecting factors is finished. An obtained image comprehensive quality score is relatively same with the subjective evaluation of human eyes, and furthermore a relatively high evaluation speed is realized. After the low-quality image is filtered, face identification performance can be effectively improved.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, in particular to a method and system for evaluating the quality of human face images based on multi-dimensional evaluation standards. Background technique [0002] With the rapid development of society and technology, face recognition technology has been more and more widely used and popularized, such as airports, railway stations and other important security ports, banks, communities and other important places closely related to people's lives , these applications have higher requirements on the accuracy and efficiency of face recognition algorithms. In the face recognition algorithm, the training data input by the model is one of the most important factors affecting the performance of the algorithm, so the quality of the training input image has a great direct impact on the performance of the model. In the process of face image data collection, due to camera shake, posture, occlu...

AI Technical Summary

Problems solved by technology

There are reference evaluation results that are closer to subjective evaluation results, but this type of method is extremely dependent on reference standard face images, and in actual face recognition applications, it is generally impossible or difficult to obtain high-quality standard face images, so The no-reference evaluation method has become a hot and difficult issue in the field of image quality evaluation

Images