isp 图像算法(二)之dead pixel correction坏点矫正

代码在git

相机中的坏点就是那些和周围不一样的点，就是那些数值极大或者极小值点，你可以理解一张曲面的山峰或者山谷，人群中也是一样，那些与大众不一样的人就是"坏人",衡量好坏用他与周围的差值，

abs[V(好人)-v(坏人)]

 if (abs(p1 - p0) > self.thres) and (abs(p2 - p0) > self.thres) and (abs(p3 - p0) > self.thres) \
                        and (abs(p4 - p0) > self.thres) and (abs(p5 - p0) > self.thres) and (abs(p6 - p0) > self.thres) \
                        and (abs(p7 - p0) > self.thres) and (abs(p8 - p0) > self.thres):

threds=30

那个最优秀的人就是坏人,不，是坏pixel

p6	p7	p8
p4	p0	p5
p1	p2	p3

代码

#!/usr/bin/python
import numpy as np

class DPC:
    'Dead Pixel Correction'

    def __init__(self, img, thres, mode, clip):
        self.img = img
        self.thres = thres
        self.mode = mode
        self.clip = clip

    def padding(self):
        #在四周放两个0 从(1080,1920) --->(1084,1924)
        img_pad = np.pad(self.img, (2, 2), 'reflect')
        return img_pad

    def clipping(self):
        
        #np.clip是一个截取函数，用于截取数组中小于或者大于某值的部分，并使得被截取部分等于固定值
        #限定在()0,1023
        np.clip(self.img, 0, self.clip, out=self.img)
        return self.img

    def execute(self):
        img_pad = self.padding()
        raw_h = self.img.shape[0]
        raw_w = self.img.shape[1]
        dpc_img = np.empty((raw_h, raw_w), np.uint16)
        for y in range(img_pad.shape[0] - 4):
            for x in range(img_pad.shape[1] - 4):
                p0 = img_pad[y + 2, x + 2]
                p1 = img_pad[y, x]
                p2 = img_pad[y, x + 2]
                p3 = img_pad[y, x + 4]
                p4 = img_pad[y + 2, x]
                p5 = img_pad[y + 2, x + 4]
                p6 = img_pad[y + 4, x]
                p7 = img_pad[y + 4, x + 2]
                p8 = img_pad[y + 4, x + 4]
                if (abs(p1 - p0) > self.thres) and (abs(p2 - p0) > self.thres) and (abs(p3 - p0) > self.thres) \
                        and (abs(p4 - p0) > self.thres) and (abs(p5 - p0) > self.thres) and (abs(p6 - p0) > self.thres) \
                        and (abs(p7 - p0) > self.thres) and (abs(p8 - p0) > self.thres):
                    if self.mode == 'mean':
                        p0 = (p2 + p4 + p5 + p7) / 4
                    elif self.mode == 'gradient':
                        dv = abs(2 * p0 - p2 - p7)
                        dh = abs(2 * p0 - p4 - p5)
                        ddl = abs(2 * p0 - p1 - p8)
                        ddr = abs(2 * p0 - p3 - p6)
                        if (min(dv, dh, ddl, ddr) == dv):
                            p0 = (p2 + p7 + 1) / 2
                        elif (min(dv, dh, ddl, ddr) == dh):
                            p0 = (p4 + p5 + 1) / 2
                        elif (min(dv, dh, ddl, ddr) == ddl):
                            p0 = (p1 + p8 + 1) / 2
                        else:
                            p0 = (p3 + p6 + 1) / 2
                dpc_img[y, x] = p0
        self.img = dpc_img
        return self.clipping()