python实现图像检索的三种(直方图/OpenCV/哈希法)

#统计直方图，用load()载入图片的像素pix，再分别读取每个像素点的R\G\B值进行统计（分别为0-255）
#将256个颜色值的统计情况投影到32个，返回R\G\B投影后的统计值数组，共32*3=96个元素
def calc_Hist(img):
  '''
  #120张图片，4.43s
  w,h = img.size
  pix = img.load() #载入图片，pix存的是像素
  calcR = [0 for i in range(0,32)]
  calcG = [0 for i in range(0,32)]
  calcB = [0 for i in range(0,32)]
  for i in range(0,w):
    for j in range(0,h):
      (r,g,b) = pix[i,j]
      #print (r,g,b)
      calcR[r/8] += 1
      calcG[g/8] += 1
      calcB[b/8] += 1
  calcG.extend(calcB)
  calcR.extend(calcG)

  return calcR
  '''
  #120张图，3.49s

  w,h = img.size
  pix = img.load() #载入图片，pix存的是像素
  calcR = [0 for i in range(0,256)]
  calcG = [0 for i in range(0,256)]
  calcB = [0 for i in range(0,256)]
  for i in range(0,w):
    for j in range(0,h):
      (r,g,b) = pix[i,j]
      #print (r,g,b)
      calcR[r] += 1
      calcG[g] += 1
      calcB[b] += 1
  calcG.extend(calcB)
  calcR.extend(calcG) #256*3

  #calc存放最终结果，32*3
  calc = [0 for i in range(0,96)]
  step = 0 #calc的下标，0~95
  start = 0 #每次统计的开始位置
  while step < 96:
    for i in range(start,start+8): #8个值为1组，统计值相加，eg：色彩值为0~7的统计值全部转换为色彩值为0的统计值
      calc[step] += calcR[i]
    start = start+8
    step += 1
  #print calc 
  return calc 

#该函数用于统一图片大小为256*256，并且分割为16个块，返回值是16个局部图像句柄的数组
def split_Img(img, size = (64,64)):
  img = img.resize((256,256)).convert('RGB')
  w,h = img.size
  sw,sh = size
  return [img.crop((i,j,i+sw,j+sh)).copy() for i in xrange(0,w,sw) for j in xrange(0,h,sh)]

#计算两个直方图之间的相似度，h2和h3为直方图，zip表示同步遍历
def calc_Similar(h2,h3):
  return sum(1 - (0 if g==s else float(abs(g-s))/max(g,s)) for g,s in zip(h2,h3)) / len(h2)

results = {} #记录结果
reverse = True #correlation/intersection方法reverse为true，另外两种为false

imgCV = cv2.imread(self.testImg.encode('utf-8'))
#self.testImg为待匹配图片
testHist = cv2.calcHist([imgCV],[0,1,2],None,[8,8,8],[0,256,0,256,0,256])
#提取直方图
testHist = cv2.normalize(testHist,testHist,0,255,cv2.NORM_MINMAX).flatten()
#均衡化

#计算self.testImg和其他图片的直方图差异，INTERSECTION方法效果比较好
for (k, hist) in self.index_cv.items(): 
#self.index_cv保存的是图片库中图片的直方图信息
  d = cv2.compareHist(testHist,hist, cv2.cv.CV_COMP_INTERSECT)
  results[k] = d
  #对结果排序，以v即上面的d作为关键字
  results = sorted([(v, k) for (k, v) in results.items()], reverse = reverse) 
  end = time.time()
  print 'OpenCV Time:'
  print end-start     
self.DisplayTotalPics(results)

#获取排序时的关键值（即汉明距离）    
def getKey(x): 
  return int(x[1])

#由灰度图得到2值“指纹”，从而计算汉明距离
def getCode(img):
  w,h = img.size
  pixel = []
  for i in range(0,w):
    for j in range(0,h):
      pixel_value = img.getpixel((i,j))
      pixel.append(pixel_value) #加入pixel数组
  avg = sum(pixel)/len(pixel) #计算像素平均值

  cp = [] #二值数组
  for px in pixel:
    if px > avg:
      cp.append(1)
    else:
      cp.append(0)
  return cp

#计算两个编码之间的汉明距离
def compCode(code1,code2):
  num = 0
  for index in range(0,len(code1)):
    if code1[index] != code2[index]:
      num+=1
  #print num
  #print '\n'
  return num