Panoramic image fusion method based on depth convolution neural network and depth information

Abstract

The invention discloses a panoramic image fusion method based on a depth convolution neural network and depth information. The method comprises the steps of (S1) constructing a deep learning training data set, selecting overlap regions xe1 and xe2 of two fish eye images to be fused used for training and an ideal fusion area ye of a panoramic image formed after fusing the two fish eye images, and constructing a training set {xe1, xe2, ye} of the images to be fused and a panoramic image block pair, (S2) constructing a convolution neural network model, and (S3) obtaining a fusion area of a test data set based on a test data set and a trained depth convolution neural network model. According to the method, an image can be expressed more comprehensively and deeply, the image semantic representation in a plurality of abstract levels is realized, and the accuracy of image fusion is improved.

Description

technical field [0001] The invention belongs to the technical field of image communication, relates to the technical field of image splicing, and in particular relates to a panoramic image fusion method based on a deep convolutional neural network and depth information. Background technique [0002] Image stitching technology is the technology of stitching several partially overlapping images into a large seamless high-resolution image. Use ordinary cameras to obtain wide-field scene images. Because the resolution of the camera is fixed, the larger the scene is, the lower the image resolution will be. However, panoramic cameras and wide-angle lenses are not only very expensive, but also have serious distortion. In order to obtain an ultra-wide field of view or even a 360-degree panorama without reducing the image resolution, a computer image stitching method has emerged. [0003] Image stitching is one of the key technologies in image processing, and it is the basis for oth...

AI Technical Summary

Problems solved by technology

Currently, there are two types of commonly used image fusion methods. One is to use direct fusion methods (for example: average value method, weighted average method, median filter method), which will cause the generated panoramic image to be blurred due to the difference in details in the overlapping area. There are obvious stitching seams; the other is to use dynamic programming and graph-cut methods to find the optimal fusion centerline, specifically, to use the grayscale difference and color difference between pixels in the overlapping area of ​​the two images, Find a line with the smallest grayscale and color difference in the overlapping area, and then select a seam width on the left and right sides of the optimal fusion center line for linear fusion, which will cause ghosting and seams in the image fusion area; therefore This area urgently needs a kind of panoramic image mosaic method that can overcome above-mentioned defective

Images