All-focus image generation method based on interactive adversarial learning

Abstract

The invention belongs to the technical field of image information processing, and provides a full-focus image generation method based on interactive adversarial learning. According to the full-focus image generation method, an interactive adversarial learning mode is utilized, firstly, the complementary relation between a non-focus fuzzy detection task and an edge detection task is utilized, and a boundary focused to a non-focus transition area and a homogeneous area with little detection texture information are better located; secondly, paired non-focusing images and corresponding full-focusing true-value images are needed in most of the current common methods, and the paired images are difficult to obtain. However, the problem does not exist in the method disclosed by the invention, and the natural all-focus image generation network with global consistency can be realized only by using all-focus images which are not paired as references.

Description

technical field [0001] The invention belongs to the technical field of image information processing, in particular to a method for generating an all-focus image. Background technique [0002] At present, the methods related to this patent include two aspects: the first is an out-of-focus blur detection algorithm; the second is an image generation algorithm based on generative confrontation. [0003] Non-focus blur detection methods are mainly divided into two categories: Traditional methods based on artificial design, which mostly extract features through manual design, usually use image gradient, frequency and other features to construct detectors to complete non-focus blur area detection . In the document "Spatially-varying blur detection based on multiscale fused and sorted transform coefficients of gradient magnitudes", Golestaneh et al. proposed a spatially varying blur detection method based on high-frequency multi-scale fusion and gradient size sorting transformation...

AI Technical Summary

Problems solved by technology

In the document "Spatially-varying blur detection based on multiscale fused and sorted transform coefficients of gradient magnitudes", Golestaneh et al. proposed a spatially varying blur detection method based on high-frequency multi-scale fusion and gradient size sorting transformation, and performed local calculations at each pixel. , to determine the fuzzy level, and set the parameters to build the model by manually designing the feature extraction method, so the complexity of the model is low, but there are also problems that it is difficult to extract deep features and poor robustness; deep learning based on convolutional neural network method, in these methods, most of them use the source image as input, use the non-focus area detection truth value as supervision, and use information such as multi-scale or multi-level feature fusion to obtain the final detection result map

Images