A Fast Automatic White Balance and Color Correction Method Based on Deep Learning

Abstract

The present invention provides a fast automatic white balance and color correction method based on deep learning. In terms of speed: awb and color correction can achieve real-time performance on ordinary embedded devices, and the amount of calculation is reduced compared with traditional algorithms; in terms of effect: solve It solves the problem of inaccurate estimation of white balance of large-area solid-color images by traditional methods, and shows better white balance and color correction effects in low color temperature and low illumination environments. After the imatest test, under low illumination and low color temperature, The gray scale reduction degree of white balance is above 90%, the ΔC<8% and ΔE<15% after color correction, and the white balance reduction degree under high color temperature is over 95%, and the color saturation is basically greater than 100. However, the traditional white balance algorithm can only achieve 60%-80% grayscale reduction under low color temperature and low illumination. In terms of applicability: this method is suitable for multiple sensors and can be extended to other task fields.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a fast automatic white balance and color correction method based on deep learning. Background technique [0002] The traditional AWB algorithm includes some of the following algorithms: [0003] Advanced white balance algorithm, its algorithm flow is: [0004] 1. Take gray card pictures in different color temperatures (D75, D65, D50, CWF, H, A) in the light box, the gray card is full of the whole set of live pictures; [0005] 2. Calibrate the white balance reference point at each color temperature, and then determine the reference white point to draw the reference white area; [0006] 3. Calculate the reference white point where the image to be corrected currently falls into the reference white area, and calculate the white balance color compensation of the image to be corrected according to these gray points combined with the weight; [0007] Gray world ...

AI Technical Summary

Problems solved by technology

[0028] The main problem of the advanced white balance algorithm is that there is inaccurate detection of the gray area of ​​the white point, and there will be wrong white point detection for the detection of the gray area of ​​the large-area solid color background, which will eventually be corrected into a white balance image with a color cast

[0029] The main problem with the grayscale world method is that if the color of the image is relatively single, then the grayworld assumption is no longer satisfied.

[0030] The main problem of the perfect reflection method is that it is greatly affected by the parameter of the threshold T

[0035] For these neural network tasks, it is difficult to calculate white balance and color correciton parameters in real time and it is not practical for multiple sensors

Images