double_jpeg_compression.py

import numpy as np
# import pandas as pd
import cv2
# import argparse
# import csv
# import sys

from scipy import fftpack as fftp
from matplotlib import pyplot as plt


def detect(image):
    firstq = 30
    secondq = 40
    thres = 0.5

    dct_rows = 0
    dct_cols = 0

    image = cv2.imread(image)
    # image = cv2.imread(image)
    shape = image.shape

    if shape[0] % 8 != 0:
        dct_rows = shape[0]+8-shape[0] % 8
    else:
        dct_rows = shape[0]

    if shape[1] % 8 != 0:
        dct_cols = shape[1]+8-shape[1] % 8
    else:
        dct_cols = shape[1]

    dct_image = np.zeros((dct_rows, dct_cols, 3), np.uint8)
    dct_image[0:shape[0], 0:shape[1]] = image

    y = cv2.cvtColor(dct_image, cv2.COLOR_BGR2YCR_CB)[:, :, 0]

    w = y.shape[1]
    h = y.shape[0]
    n = w*h/64

    Y = y.reshape(h//8, 8, -1, 8).swapaxes(1, 2).reshape(-1, 8, 8)

    qDCT = []

    for i in range(0, Y.shape[0]):
        qDCT.append(cv2.dct(np.float32(Y[i])))

    qDCT = np.asarray(qDCT, dtype=np.float32)
    qDCT = np.rint(qDCT - np.mean(qDCT, axis=0)).astype(np.int32)
    f, a1 = plt.subplots(8, 8)
    a1 = a1.ravel()

    k = 0
    # flag = True
    for idx, ax in enumerate(a1):
        k += 1
        data = qDCT[:, int(idx/8), int(idx % 8)]
        val, key = np.histogram(data, bins=np.arange(data.min(), data.max()+1))
        # val, key = np.histogram(data, bins=np.arange(data.min(), data.max()+1), normed=True)
        z = np.absolute(fftp.fft(val))
        z = np.reshape(z, (len(z), 1))
        rotz = np.roll(z, int(len(z)/2))

        slope = rotz[1:] - rotz[:-1]
        indices = [i+1 for i in range(len(slope)-1)
                   if slope[i] > 0 and slope[i+1] < 0]

        peak_count = 0

        for j in indices:
            if rotz[j][0] > thres:
                peak_count += 1

        if(k==3):
            if peak_count>=20: return True
            else: return False
            # flag = False