scbc_repos/ssat_sdk/picture/pq_detect.py

# -*- coding:utf-8 -*-
from RGB import *
from image_util import *

import os, sys, time
import cv2 as cv
import numpy as np
import random

class PQDetect():
    def __init__(self):
        self.RGBColor = RGBColor()

    def isSnowPic2(self, pic_path):
        im1 = cv.imread(pic_path)
        h1, w1, c1 = im1.shape
        print h1, w1, c1
        im1_1 = im1[0:h1 / 2, 0:w1 / 2]
        im1_2 = im1[0:h1 / 2, w1 / 2:w1]
        im1_3 = im1[h1 / 2:h1, 0:w1 / 2]
        im1_4 = im1[h1 / 2:h1, w1 / 2:w1]
        hist_list = []
        hist_list.append(cv.calcHist([im1_1], [0], None, [256], [0, 256]))
        hist_list.append(cv.calcHist([im1_2], [0], None, [256], [0, 256]))
        hist_list.append(cv.calcHist([im1_3], [0], None, [256], [0, 256]))
        hist_list.append(cv.calcHist([im1_4], [0], None, [256], [0, 256]))
        count = 0
        totalHist = 0
        for hist_1 in hist_list:
            for hist_2 in hist_list:
                diffHist = cv.compareHist(hist_1, hist_2, cv.HISTCMP_BHATTACHARYYA)
                if diffHist <> 0:
                    count += 1
                    totalHist += diffHist
        four_zone_diff = totalHist / count
        print four_zone_diff

        rhist = cv.calcHist([im1_1], [0], None, [256], [0, 256])
        bhist = cv.calcHist([im1_2], [2], None, [256], [0, 256])
        bgr_diff = cv.compareHist(rhist, bhist, cv.HISTCMP_BHATTACHARYYA)
        print bgr_diff

        if four_zone_diff < 0.1 and bgr_diff < 0.2:
            return True
        else:
            return False

    '''
    判断传入的图片，是否为ATV无信号雪花屏。定义雪花屏：清晰度25左右.然后随机取20个区域，查看直方图趋势相似度。
    :param pic_path:图片路径
    :threshold： 直方图差异临界值。 范围 0 ~ 1
    :sharpDiff: 清晰度容错范围值。
    :return : True代表是雪花屏，False代表不是雪花屏
    '''
    def isSnowPic(self, pic_path, threshold=0.1, sharpDiff=10, sharpSnow = 25):
        im1 = cv.imread(pic_path)
        h1, w1, c1 = im1.shape

        sharpP = self.detSharpLaplacian(pic_path)

        hist_list = []
        for i in range(20):
            x1 = random.randint(0,h1/2)
            y1 = random.randint(0,w1/2)
            rect_h = h1/2
            rect_w = w1/2
            # print [x1,y1, x1+ rect_w, y1+rect_h]
            im = cutMat(im1, [x1,y1, x1+ rect_h, y1+rect_w])
            hist = cv.calcHist([im], [0], None, [256], [0, 256])
            hist_list.append(hist)

        count = 0
        totalHist = 0
        for hist_1 in hist_list:
            for hist_2 in hist_list:
                diffHist = cv.compareHist(hist_1, hist_2, cv.HISTCMP_BHATTACHARYYA)
                if diffHist <> 0:
                    count += 1
                    totalHist += diffHist
        four_zone_diff = totalHist / count

        print "isSnowPic:", four_zone_diff,sharpP

        if four_zone_diff < threshold and abs(sharpP - sharpSnow) <=  sharpDiff:
            return True
        else:
            return False

    '''
    根据传入的图片路径，计算图片色温
    '''
    def calculateCCT(self, pic_path):
        img = cv.imread(pic_path)
        xyImg = cv.cvtColor(img,cv.COLOR_BGR2XYZ)
        #色温计算
        X = np.average(xyImg[:,:,0])
        Y = np.average(xyImg[:,:,1])
        Z = np.average(xyImg[:,:,2])
        x = X/(X+Y+Z) #色谱XYZ转换成色坐标x
        y = Y/(X+Y+Z) #色谱XYZ转换成色坐标y
        print "calculateCCT,x,y:",x,y
        n = (x-0.3320)/(0.1858-y)
        print "calculateCCT,n:",n
        CCT = 437*n**3 + 3601*n**2 + 6831*n + 5517
        print "calculateCCT,CCT:",CCT
        return CCT

    '''
        Tenengrad梯度方法利用Sobel算子分别计算水平和垂直方向的梯度，同一场景下梯度值越高，图像越清晰。
        以下是具体实现，这里衡量的指标是经过Sobel算子处理后的图像的平均灰度值，值越大，代表图像越清晰。
        :param 图片路径
        :return float， 值越大，代表清晰度越高
        '''

    def detSharpTenengrad(self, pic_path):
        img = cv.imread(pic_path)
        grayImg = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

        sobelImg = cv.Sobel(grayImg, cv.CV_16U, 1, 1)
        grayV = cv.mean(sobelImg)[0]  # 图像的平均灰度

        # cv.putText(img, str(grayV), (20,50), cv.FONT_HERSHEY_COMPLEX, 0.8, (255,255,0))
        # cv.imshow("清晰度",img)
        # cv.waitKey(0)
        return grayV

    '''
    采用Laplacian梯度方法检测清晰度
    :param 图片路径
    :return float， 值越大，代表清晰度越高
    '''

    def detSharpLaplacian(self, pic_path):
        img = cv.imread(pic_path)
        return self.detImgSharpLaplacian(img)

    def detImgSharpLaplacian(self, img):
        grayImg = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

        sobelImg = cv.Laplacian(grayImg, cv.CV_16U)[0]
        grayV = cv.mean(sobelImg)[0]  # 图像的平均灰度

        # cv.putText(img, str(grayV), (20, 50), cv.FONT_HERSHEY_COMPLEX, 0.8, (255, 255, 0))
        # cv.imshow("清晰度", img)
        # cv.waitKey(0)
        return grayV

    '''
    采用方差（Variance）方法检测清晰度。
    方差是概率论中用来考察一组离散数据和其期望（即数据的均值）之间的离散（偏离）成都的度量方法。
    方差较大，表示这一组数据之间的偏差就较大，组内的数据有的较大，有的较小，分布不均衡；
    方差较小，表示这一组数据之间的偏差较小，组内的数据之间分布平均，大小相近。
    :param 图片路径
    :return float， 值越大，代表清晰度越高
    '''

    def detSharpVariance(self, pic_path):
        img = cv.imread(pic_path)
        grayImg = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

        # 求灰度图像的标准差
        stdValueImg = cv.meanStdDev(grayImg)
        # print stdValueImg
        meanValue = stdValueImg[1][0][0]  # 求得均方差。stdValueImg：(均差array([[ 16.58646894]]), 标准差array([[ 10.1834467]]))
        grayV = meanValue * meanValue
        # cv.putText(img, str(grayV), (20, 50), cv.FONT_HERSHEY_COMPLEX, 0.8, (255, 255, 0))
        # cv.imshow("清晰度", img)
        # cv.waitKey(0)
        return grayV