region-based color histogram feature.

sherlockchou86/cbir-image-search/cbir_ch_feature.py

import numpy as np
import cv2

#HSV 颜色空间中每个通道的 bins,可进行调整
bins = (8, 12, 3)

#计算单个区域的颜色直方图特征
def histogram(image, mask):
    #根据设定的 bins，从图像的 mask 区域提取 3D 颜色直方图.
    hist = cv2.calcHist([image], [0, 1, 2], mask, bins, [0, 180, 0, 256, 0, 256])

    #直方图归一化
    hist = cv2.normalize(hist, hist).flatten()

    return hist

#计算基于区域颜色直方图特征，特征向量维度为：bins[0]*bins[1]*bins[2]*5
def describe(img_cv2):
    #转换到 HSV 空间.
    img_hsv = cv2.cvtColor(img_cv2, cv2.COLOR_BGR2HSV)
    h, w = img_hsv.shape[:2]
    cX, cY = (int(w * 0.5), int(h * 0.5)) #图片中心
      
    #将图像化分为 4 块：
    #(top-left, top-right, bottom-right, bottom-left)
    segments = [(0, cX, 0, cY), 
                (cX, w, 0, cY), 
                (cX, w, cY, h),
                (0, cX, cY, h)]
      
    #构建椭圆mask(elliptical mask) 来表示图片中心
    axesX, axesY = (int(w * 0.75) // 2, int(h * 0.75) // 2)
    ellipMask = np.zeros(img_hsv.shape[:2], dtype = "uint8")
    cv2.ellipse(ellipMask, (cX, cY), (axesX, axesY), 0, 0, 360, 255, -1)

    #计算区域颜色直方图
    features = []
    for (startX, endX, startY, endY) in segments:
        #构建图片每个角的 mask，并提取其椭圆中心.
        cornerMask = np.zeros(img_hsv.shape[:2], dtype = "uint8")
        cv2.rectangle(cornerMask, (startX, startY), (endX, endY), 255, -1)
        cornerMask = cv2.subtract(cornerMask, ellipMask)
        
        #构建颜色直方图
        hist = histogram(img_hsv, cornerMask)
        features.extend(hist)

    #提取各椭圆区域的颜色直方图
    hist = histogram(img_hsv, ellipMask)
    features.extend(hist)
    
    #特征向量features包含 4 个角区域和 1 个中心区域五部分，
    #特征向量维度等于 bins[0]*bins[1]*bins[2]*5
    return features

#
img_cv2 = cv2.imread('test.jpg')
extract_feats = describe(img_cv2)
print("[INFO]Image color histogram feature:{}".format(extract_feats))