python3 Circular-LBP(圆形局部二值化算法)的实现及批量处理

python3 Circular-LBP(圆形局部二值化算法)的实现及批量处理

• Circular-LBP算法的实现
  由于Circula-LBP算法较传统的LBP特征提取效果更好，近期实验需要，所以用python3实现了Circular-LBP，代码如下：

import numpy as np
import cv2
from matplotlib import pyplot as plt
import math

#双极性插值
def bilinear_interpolation(x, y, img):
    x1, y1 = int(r), int(c)
    x2, y2 = math.ceil(r), math.ceil(c)

    r1 = (x2 - x) / (x2 - x1) * get_pixel_else_0(img, x1, y1) + (x - x1) / (x2 - x1) * get_pixel_else_0(img, x2, y1)
    r2 = (x2 - x) / (x2 - x1) * get_pixel_else_0(img, x1, y2) + (x - x1) / (x2 - x1) * get_pixel_else_0(img, x2, y2)

    return (y2 - y) / (y2 - y1) * r1 + (y - y1) / (y2 - y1) * r2    

#阈值设置
def thresholded(center, pixels):
    out = []
    for a in pixels:
        if a >= center:
            out.append(1)
        else:
            out.append(0)
    return out

#像素返回
def get_pixel_else_0(l, idx, idy):
    if idx < int(len(l)) - 1 and idy < len(l[0]):
        return l[idx,idy]
    else:
        return 0

#读取图像
img = cv2.imread('C:/Users/qgl/Desktop/articles/test1.jpg', 0)
transformed_img = cv2.imread('C:/Users/qgl/Desktop/articles/test1.jpg', 0)

#相邻像素P和半径R
P = 8 # number of pixels
R = 1 # radius 

for x in range(0, len(img)):
    for y in range(0, len(img[0])):
        center        = img[x,y]
        pixels = []
        for point in range(0, P):
            r = x + R * math.cos(2 * math.pi * point / P)
            c = y - R * math.sin(2 * math.pi * point / P)
            if r < 0 or c < 0:
                pixels.append(0)
                continue            
            if int(r) == r:
                if int(c) != c:
                    c1 = int(c)
                    c2 = math.ceil(c)
                    w1 = (c2 - c) / (c2 - c1)
                    w2 = (c - c1) / (c2 - c1)
                                    
                    pixels.append(int((w1 * get_pixel_else_0(img, int(r), int(c)) + \
                                   w2 * get_pixel_else_0(img, int(r), math.ceil(c))) / (w1 + w2)))
                else:
                    pixels.append(get_pixel_else_0(img, int(r), int(c)))
            elif int(c) == c:
                r1 = int(r)
                r2 = math.ceil(r)
                w1 = (r2 - r) / (r2 - r1)
                w2 = (r - r1) / (r2 - r1)                
                pixels.append((w1 * get_pixel_else_0(img, int(r), int(c)) + \
                               w2 * get_pixel_else_0(img, math.ceil(r), int(c))) / (w1 + w2))
            else:
                pixels.append(bilinear_interpolation(r, c, img))

                
        values = thresholded(center, pixels)
        res = 0
        for a in range(0, len(values)):
            res += values[a] * (2 ** a)

        transformed_img.itemset((x,y), res)
         
print (x)

cv2.imshow('image', img)
cv2.imshow('thresholded image', transformed_img)
cv2.imwrite('C:...'+'/'+'1.jpg',transformed_img)

• Circular-LBP的批量处理
  关于批量处理，需要修改之前算法中LBP函数的输出部分，转为return：

return transformed_img

再重新def定义一个循环函数，对目标路径下的每一张图像进行处理：

def tran(src,drc,P,R):
    list = os.listdir(src)#遍历数据集的所有图片
    sum = 0
    for i in list:
        try:
            img = cv2.imread(src+'/'+i,0)#按list的顺序读取每一张图片
            transformed_img = cv2.imread(src+'/'+i,0)#按list的顺序读取每一张图片

            cv2.imshow('img',img)#显示图片
            transformed_img = LBP(img,P,R,transformed_img)#调用下方定义的LBP（）函数，得到LBP计算后的图片
            cv2.imwrite(drc+'/'+i,transformed_img)#将得到的每张图片写入新的目标文件夹里
            
            
            
            sum = int(sum)+1
            print(i+'is finished, number is'+str(sum))
        except:
            print('error in'+i)

对数据集进行处理，就可以进行后续的分类工作了。