Depth learning based image quality improving method and device

Abstract

The invention provides a depth learning based image quality improving method and device. The method includes obtaining a distorted image of a scene as a training value of a depth learning model; obtaining a panoramic image of the scene as a true value of the depth learning model; learning and generating the mapping relation between the training value and the true value by utilizing an establishedfirst convolutional neural network model; selecting the distorted images of other scenes as the input of the mapping relation so that the output panoramic images of other scenes can be obtained, and comparing the panoramic images with the preset panoramic images of other scenes; and repeatedly executing training steps until the difference value generated by comparison is less than or equal to a preset threshold value, and determining the mapping relation obtained by training at the moment as a final model used for improving the imaging quality of the panoramic images. Pixel imbalance problemsin a process of stitching and generating the panoramic images can be solved, and thus, the imaging quality of images can be greatly improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image quality improvement method and device based on deep learning. Background technique [0002] The realization process of panorama (Panoramic image) technology is usually as follows: first, use camera equipment to shoot the scene at different angles or positions, and obtain multiple images reflecting the information of the entire three-dimensional scene; Zhang big picture, all the information of the whole scene is displayed on this big picture, so that the information of the whole scene can be concentrated on one picture. [0003] The current panoramic images are generally completed by splicing images taken by multiple fisheye cameras or images taken by other types of cameras, and the edge imaging quality is affected by the projection mapping effect of the original fisheye images or other types of images. When the imaging of the camera model on the image plane is not unifor...

AI Technical Summary

Problems solved by technology

[0004] In view of the shortcomings of the prior art described above, the object of the present invention is to provide a deep learning-based image quality improvement method and its device, which are used to solve the poor imaging quality of panoramic images stitched from at least one image in the prior art , unbalanced pixels, etc.

Images