Opencv removes highlights

The code is directly borrowedOpencv highlight removal_SkyrimDragonBorn's blog-CSDN blog

The original blog is in C++, I just used python to translate it.

The reference paper should be this:

（1）

 Real-Time Specular Highlight Removal Using Bilateral Filtering | SpringerLinkIn this paper, we propose a simple but effective specular highlight removal method using a single input image. Our method is based on a key observation - the maximum fraction of the diffuse color component (so called maximum diffuse chromaticity in the literature) in...https://link.springer.com/chapter/10.1007/978-3-642-15561-1_7

（2）Efficient and Robust Specular Highlight Removal | IEEE Journals & Magazine | IEEE Xplorehttps://ieeexplore.ieee.org/abstract/document/6915741

import cv2
import numpy as np

img = cv2.imread('xxx.jpg')


def highlight_remove(img):
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB).astype(np.float32)#一定要转换成float32，否则会出错
    # img = cv2.resize(img, (512, 512)).astype(np.float32)#如图片分辨率较大，直接resize到512会比较快
    dst = img.copy()
    h, w = img.shape[:2]
    for i in range(h):
        for j in range(w):
            R = img[i, j, 0]
            G = img[i, j, 1]
            B = img[i, j, 2]
            alpha_r = R/(R+G+B)
            alpha_g = G/(R+G+B)
            alpha_b = B/(R+G+B)
            # alpha = max(max(alpha_r, alpha_g), alpha_b)
            MaxC = max(max(R, G), B)
            minalpha = min(min(alpha_r, alpha_g), alpha_b)
            gama_r = (alpha_r - minalpha) / (1 - 3 * minalpha)
            gama_g = (alpha_g - minalpha) / (1 - 3 * minalpha)
            gama_b = (alpha_b - minalpha) / (1 - 3 * minalpha)
            gama = max(max(gama_r, gama_g), gama_b)

            temp = (gama * (R + G + B) - MaxC) / (3 * gama - 1)
            dst[i, j, 0] = np.clip(R-(temp+0.5), 0, 255)
            dst[i, j, 1] = np.clip(G-(temp+0.5), 0, 255)
            dst[i, j, 2] = np.clip(B-(temp+0.5), 0, 255)
    return dst[:, :, ::-1]


res = highlight_remove(img)
res = np.clip(res, 0, 255)
cv2.imwrite('xxx_remove.jpg', res)

The effect is as follows, which basically corresponds to the original c++ version. 

 However, in personal testing, this algorithm is only suitable for pictures where the original image is dark. I tried several daily photos like this and basically failed, so the scope of application of this algorithm is very limited.