Panoramic picture no-reference quality evaluation method, system and device

Abstract

The invention provides a panoramic picture no-reference quality evaluation method. The panoramic picture no-reference quality evaluation method comprises the steps of mapping a distorted picture D toa cubic projection picture from an equidistant columnar projection picture; extracting feature values of a single cubic projection view by adopting a single-channel hybrid convolutional neural networkchannel, enabling six views of the cubic projection view to respectively pass through the corresponding single-channel hybrid convolutional neural network, and then fusing the feature values extracted by the six channels together in a series mode; and regressing the features extracted by the multi-channel convolutional network to an image quality score by using a full connection layer. The invention also provides a panoramic picture no-reference quality evaluation system. According to the method, the problem of panoramic picture distortion is solved by using a cubic projection mode, and the features beneficial to quality evaluation are effectively extracted by using the hybrid convolutional neural network.

Description

technical field [0001] The present invention relates to the technical field of image quality evaluation, in particular to a method, system and device for evaluating the quality of a panoramic image without reference based on a multi-channel hybrid convolutional neural network. Background technique [0002] With the rapid development of VR technology, more and more people can watch panoramic videos and pictures through various VR devices such as Googleglass, Gear VR, HTV VIVE, etc. At the same time, panoramic videos and pictures can give people a sense of body. The feeling of being on the scene makes panoramic videos and pictures widely used in VR movies, concerts and sports live broadcasts and other occasions. However, since panoramic videos and pictures need to record 360-degree all-round picture information, the resolution of panoramic videos and pictures is generally very high, and the occupied storage space and transmission bandwidth are also large. Therefore, in real a...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the lack of effective methods for panorama picture quality assessment in the field of image quality assessment in the prior art, the purpose of the present invention is to provide a panorama picture no-reference quality assessment method, system and equipment, the method, system and equipment based on multi-channel Hybrid convolutional neural network can effectively evaluate the quality of panoramic pictures, which is of great significance to the development of panoramic pictures

Images