OpenCV—python 基于反向投影的显著性检测

我们在获取显著性检测时，我们希望进行无监督提取或者分割目标。深度学习在目标检测与显著性检测方面已经是非常成熟了，但是算法复杂度相对于传统无监督的算法来说还是高了点，这里只是写个示例。仅供参考。

其中：
关于均值漂移 cv2.pyrMeanShiftFiltering

import cv2
import numpy as np
import matplotlib.pyplot as plt


def show_calculate(name, img_hist):
  plt.plot(img_hist)
  plt.title(name)
  plt.xlabel('Bins')
  plt.ylabel('Pixels')
  plt.show()


# 反向投影
def backproject(source, levels=2, scale=1):
  hsv = cv2.cvtColor(source,  cv2.COLOR_BGR2HSV)
  hsvt = hsv.copy()
  # cv2.imshow("hsvt",hsvt)

  roihist = cv2.calcHist([hsv], [0, 1], None, [levels, levels], [0, 180, 0, 256])
  cv2.normalize(roihist, roihist, 0, 255, cv2.NORM_MINMAX)
  # show_calculate("roihist", roihist)

  dst = cv2.calcBackProject([hsvt],[0,1],roihist,[0,180,0,256], scale)
  return dst

def saliency_by_backprojection(img):
  cv2.pyrMeanShiftFiltering(img, 2, 10, img, 4)
  backproj = np.uint8(backproject(img, levels = 2))
  # cv2.imshow("backproj",backproj)

  cv2.normalize(backproj,backproj,0,255,cv2.NORM_MINMAX)
  saliencies = [backproj, backproj, backproj]
  saliency = cv2.merge(saliencies)
  # cv2.imshow("saliency",saliency)

  cv2.pyrMeanShiftFiltering(saliency, 20, 200, saliency, 2)
  # cv2.imshow("pyrMeanShiftFiltering1", saliency)

  saliency = cv2.cvtColor(saliency, cv2.COLOR_BGR2GRAY)
  cv2.equalizeHist(saliency, saliency)
  # cv2.imshow("equalizeHist", saliency)

  saliency = 255 - saliency
  (_, saliency) = cv2.threshold(saliency, 200, 255, cv2.THRESH_BINARY)
  # cv2.imshow("threshold", saliency)
  return saliency


def refine_saliency_with_grabcut(img, saliency):
  _, contours, hierarchy = cv2.findContours(saliency, cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)
  contours = sorted(contours, key=cv2.contourArea)
  rect = cv2.boundingRect(contours[-1])

  bgdmodel = np.zeros((1, 65), np.float64)
  fgdmodel = np.zeros((1, 65), np.float64)
  saliency[np.where(saliency > 0)] = cv2.GC_FGD
  mask = saliency
  cv2.grabCut(img, mask, rect, bgdmodel, fgdmodel, 1, cv2.GC_INIT_WITH_RECT)
  mask = np.where((mask==2)|(mask==0),0,1).astype('uint8')
  return mask

def backprojection_saliency(img):
  saliency = saliency_by_backprojection(img)
  mask = refine_saliency_with_grabcut(img, saliency)
  return img * mask[:, :, np.newaxis]


if __name__ == "__main__":
  img = cv2.imread("./images/flower.jpg")
  img = cv2.resize(img, (640//2, 480//2))
  segmentation = backprojection_saliency(img)

  cv2.imshow("original", img)
  cv2.imshow("segmentation", segmentation)
  cv2.waitKey(0)
  cv2.destroyAllWindows()

反向投影直方图量化运行效果：

代码运行时效果：

原图与效果图：