Image demosaicing method based on residual feature aggregation attention blocks

Abstract

The invention discloses an image demosaicing method based on residual feature aggregation attention blocks. The method comprises the following steps: 1, constructing a green channel recovery network, inputting a raw graph sampled by a Bayer optical filter array, and outputting a reconstructed green channel graph; 2, constructing a red channel recovery network, and inputting a red sampling image and a reconstructed green channel image, wherein the output is a reconstructed red channel map; 3, constructing a blue channel recovery network; 4, fusing the reconstructed red channel image, green channel image, and blue channel image into a reconstructed RGB image; 5, calculating an average absolute error between the reconstructed RGB image and a corresponding real image according to an established network model based on the residual feature aggregation attention blocks, and optimizing the network model by taking minimization of an L1 loss function as a target. According to the method, the network can learn more features, so that the reconstruction effect is improved to obtain high-quality reconstructed images.

Description

technical field [0001] The invention belongs to the technical field of image processing and deep learning, and relates to an image demosaicing method for aggregating attention blocks based on residual features. Background technique [0002] A color image is generally expressed by three color components of red (R), green (G) and blue (B), and each color component is called a color channel. Nowadays, RGB digital cameras are the most common way to record color images, and most digital cameras use a single sensor imaging structure. The surface of the sensor is covered with a filter. The image obtained by sampling is called a raw image, and the process of reconstructing the other two color component information that is not directly sampled at each pixel position is called image demosaicing. [0003] In the current RGB digital camera, the most common color filter array is the Bayer filter array, and its imaging area is composed of 2×2 repeated arrays, and each 2×2 array contains ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide an image demosaic method based on residual feature aggregation attention block, which solves the problems of artifacts, blurring, difficult training and low reconstruction accuracy in reconstructed images when demosaicing based on deep learning in the prior art

Images