Euler video color amplification method based on deep learning

Abstract

The invention discloses an Euler video color amplification method based on deep learning. The method comprises two steps of obtaining a color amplification network and calling the color amplification network. The method comprises the first step of synthesizing a picture data set for simulating a tiny color change, and training a color amplification network constructed by a spatial decomposition module, a differential filtering module, an amplification processing module and an image reconstruction module on the data set; during operation, optimizing a FIR band-pass filter, and replacing the FIR band-pass filter with a differential filtering module of the network; and the step 2 of, when color amplification processing is carried out according to a given input video, decomposing the video into a frame sequence, generating a color amplification frame sequence by calling the color amplification network, and finally synthesizing a color amplification video. Compared with a linear amplification method, the method has the advantages that a deep learning model is used in the first step, the training process is automatic, and tedious manual design is omitted; and through processing in the step 2, noise is greatly reduced, artifacts cannot be generated in a dynamic scene, and the amplification effect is enhanced.

Description

technical field [0001] The invention relates to the field of image and video processing, in particular to a deep learning-based Euler video color amplification method. Background technique [0002] Some color signals that exist objectively but cannot be directly observed by the naked eye contain rich information, such as periodic changes in skin color with blood circulation, local skin color changes caused by slight pressing, etc. The Euler video color amplification method has an excellent effect on enhancing the color changes that are difficult to perceive by the naked eye. It can capture and amplify the tiny color signals in the video, so that it can be directly observed by the human eye, providing a better way for interpreting such information. a visualization method. [0003] But there are some deficiencies in the methods that have been realized so far. First, the spatial decomposition process of the algorithm relies on manual design, and cannot separate the color sign...

AI Technical Summary

Problems solved by technology

First of all, the spatial decomposition process of the algorithm relies on manual design, and cannot separate the color signal from the motion signal, resulting in the amplification of the color change while the motion is also amplified, which will interfere with the color amplification result

Secondly, the existing time-domain filtering methods cannot take into account both static and dynamic scenes, and there are obvious artifacts in the amplification results in dynamic scenes

Third, the amplification results often have serious noise and are prone to distortion

Images