ricezilla/video_tampering_detection

1import matplotlib.pyplot as plt2from PIL import ImageFont3from PIL import ImageDraw 4import multiprocessing5from PIL import Image6import numpy as np7import itertools8# import logging9import math10import cv211import os12 13 14# logging.basicConfig(filename=f'{os.getcwd()}/frame_processing.log', level=logging.INFO)15# logging.info('Starting frame processing')16fps = 017def read_file(name):18    global fps19    cap = cv2.VideoCapture(name)20    fps = cap.get(cv2.CAP_PROP_FPS)21    if not cap.isOpened():22        # logging.error("Cannot open Video")23        exit()24    frames = []25    while True:26        ret,frame = cap.read()27        if not ret:28            # logging.info("Can't receive frame (stream end?). Exiting ...")29            break30        frames.append(frame)31 32    cap.release()33    cv2.destroyAllWindows()34    for i in range(len(frames)):35        # print(frames[i].shape)36        frames[i]=cv2.cvtColor(frames[i], cv2.COLOR_BGR2GRAY)37 38    frames_with_index = [(frame, i) for i, frame in enumerate(frames)]39    return frames_with_index40 41st = [0,1,2,3,4]42dt = {}43idx = 0;44l = (tuple(i) for i in itertools.product(st, repeat=4) if tuple(reversed(i)) >= tuple(i))45l=list(l)46cnt = 047for i in range(0,len(l)):48    lt=l[i]49    mirror = tuple(reversed(lt))50    dt[mirror]=i;51    dt[lt]=i;52 53 54def calc_filtered_img(img):55    # residual_img= np.zeros(img.shape)56    # residual_img = np.array(img);57    fil = np.array([[-1,3,-3,1]])58    residual_img = cv2.filter2D(img, -1, fil)59    # for i in range(img.shape[0]):60    #     for j in range(img.shape[1]):61    #         residual_img[i, j] = - 3*img[i, j];62    #         if(j>0):63    #             residual_img[i, j] += img[i, j-1]64    #         if(j+1<img.shape[1]):65    #             residual_img[i, j] += 3*img[i, j+1]66    #         if(j+2<img.shape[1]):67    #             residual_img[i,j]-= img[i, j+2]68    # residual_img = np.convolve(img,[1,-3,3,-1],mode='same')69    return residual_img70 71def calc_q_t_img(img, q, t):72    # qt_img = np.zeros(img.shape)73    # for i in range(img.shape[0]):74    #     for j in range(img.shape[1]):75            # val = np.minimum(t, np.maximum(-t, np.round(img[i, j]/q)))76    #         qt_img[i, j] = val77    # print(dct)78    qt_img = np.minimum(t, np.maximum(-t, np.round(img/q)))79    return qt_img80 81def process_frame(frame_and_index):82    frame, index = frame_and_index83    # processing logic for a single frame84    # logging.info(f"Processing frame {index}")85    filtered_image = calc_filtered_img(frame)86    output_image = calc_q_t_img(filtered_image, q, t)87    output_image=output_image+288    # plt.imshow(output_image)89    return output_image.astype(np.uint8)90 91# Center the filtered image at zero by adding 12892q = 393t = 294def process_video(frames_with_index):95    num_processes = multiprocessing.cpu_count()96    # logging.info(f"Using {num_processes} processes")97    pool = multiprocessing.Pool(num_processes)  98    # process the frames in parallel99    processed_frames = pool.map(process_frame, frames_with_index)100    pool.close() 101    pool.join()102    processed_frame_with_index = [(frame, i) for i, frame in enumerate(processed_frames)]103    return processed_frame_with_index104 105co_occurrence_matrix_size = 5106co_occurrence_matrix_distance = 4107def each_frame(frame_and_index,processed_frames):108    # go rowise and column wise109    frame,index = frame_and_index110    freq_dict = {}111    for i in range( frame.shape[0]):112        for j in range( frame.shape[1]-co_occurrence_matrix_distance):113            row = frame[i]114            v1 = row[j:j+4]115            k1 = tuple(v1)116            freq_dict[k1]=freq_dict.get(k1,0)+1117    freq_dict2={}118    for i in range( frame.shape[0]-co_occurrence_matrix_distance):119        for j in range( frame.shape[1]):120            column = frame[:, j]121            v2 = column[i:i+4]122            k2 = tuple(v2)123            freq_dict2[k2]=freq_dict2.get(k2,0)+1124    freq_dict3={}125    for i in range( frame.shape[0]):126        for j in range( frame.shape[1]):127            # get next possible 4 frames128            if index < len(processed_frames)-3:129                f1 = processed_frames[index+1][i,j]130                f2 = processed_frames[index+2][i,j]131                f3 = processed_frames[index+3][i,j]132                k = (frame[i,j], f1, f2, f3)133                freq_dict3[k]=freq_dict3.get(k,0)+1134    # logging.info(f"hist made for frame {index}")135    return (freq_dict,freq_dict2,freq_dict3)136 137def extract_video(processed_frame_with_index):138    processed_frames = [frame for frame, index in processed_frame_with_index]139    num_processes = multiprocessing.cpu_count()140    # logging.info(f"Using2 {num_processes} processes")141    pool = multiprocessing.Pool(num_processes)  142    # process the frames in parallel143    freq_dict_list = pool.starmap(each_frame, zip(processed_frame_with_index,itertools.repeat(processed_frames)))144    pool.close() 145    pool.join()146    return freq_dict_list147def final(freq_dict_list):148    descriptors = []149    for freq_dicts in freq_dict_list:150        di1=[]151        for freq_dict in freq_dicts:152            frame = np.zeros(325);153            for(k,v) in freq_dict.items():154                frame[dt[k]]+=v155            di1.append(frame);156        descriptors.append(di1)157    descriptors=np.array(descriptors);158    desc_1d = descriptors.reshape(descriptors.shape[0],-1)159    mean_1d = np.mean(desc_1d,axis=0)160    co_variance_1d = np.zeros((1,1))161    for frame in desc_1d:162        mean_1d+=frame163    mean_1d=frame/len(desc_1d)164        165    for frame in desc_1d:166        tmp = frame-mean_1d167        co_variance_1d+=np.matmul(tmp,tmp.T)168    co_variance_1d=co_variance_1d/len(desc_1d)169 170    mean = np.zeros(descriptors[0].shape)171    co_variance = np.zeros((3,3))172    for frame in descriptors:173        mean+=frame174    mean=frame/len(descriptors)175 176    # print(mean)177    for frame in descriptors:178        tmp=frame-mean179        tc=np.matmul(tmp,tmp.T)180        co_variance+=tc181 182    co_variance=co_variance/len(descriptors)183    return (mean,co_variance,descriptors,mean_1d,co_variance_1d,desc_1d)184 185def final_main(input1,input2):186    f1 = read_file(input1) 187    of1 = read_file(input2)188    pf1 = process_video(f1)189    print("video1 processed residual and quantization")190    pof1=process_video(of1)191    print("video2 processed residual and quantization")192    fd1 = extract_video(pf1)193    print("video1 Created co-variance matrix")194    ofd1 = extract_video(pof1)195    print("video2 Created co-variance matrix")196    mean1,co_variance1,disc1,mean_1d_1,co_variance_1d_1,desc_1d_1=final(fd1)197    mean2,co_variance2,disc2,mean_1d_2,co_variance_1d_2,desc_1d_2=final(ofd1)198    distances = []199    print("creating Descriptors");200    for index,disc in enumerate(disc1):201        gm = disc - mean2202        dm = np.matmul(np.matmul(gm.T,np.linalg.inv(co_variance2)),gm)203        dm_sq = np.sqrt(np.abs(dm))204        distances.append(dm_sq)205 206    distances = np.array(distances)207 208    dist2 = []209    for index, disc in enumerate(disc2):210        gm = disc - mean2211        dm = np.matmul(np.matmul(gm.T,np.linalg.inv(co_variance2)),gm)212        dm_sq = np.sqrt(np.abs(dm))213        dist2.append(dm_sq)214 215    dist2 = np.array(dist2)216    fourcc = cv2.VideoWriter_fourcc(*'mp4v') 217    height =f1[0][0].shape[0]+of1[0][0].shape[0]218    width = 325+f1[0][0].shape[1]219    video = cv2.VideoWriter('video.mp4', fourcc, 30, (width,height))220    inital_diff,final_diff = 10000,-1221    result = ''222    print("writing video")223    for index, dist in enumerate(distances):224        heatmap = dist;225        frame,index = f1[index]226        different = False227        if index<len(of1):228            frame2 = of1[index][0]229            diff = dist - dist2[index]230            if not np.allclose(diff, np.zeros(diff.shape)):231                different = True232                inital_diff = min(inital_diff, index)233                final_diff = max(final_diff, index)234                sum1= np.sum(dist)235                sum2 = np.sum(dist2[index])236                    237        new_im = Image.new('RGB', (width, height))238        new_im.paste(Image.fromarray(frame), (0, 0))239        new_im.paste(Image.fromarray(frame2), (0, frame.shape[0]))240        heatmapshow = None241        heatmapshow = cv2.normalize(heatmap, heatmapshow, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)242        heatmapshow = cv2.applyColorMap(heatmapshow, cv2.COLORMAP_JET)243        new_im.paste(Image.fromarray(heatmapshow), (frame.shape[1], 0))244 245        draw = ImageDraw.Draw(new_im)246        text = "The images are same."247        if different:248            text = "The images are different."249        text_width, text_height = draw.textsize(text)250 251        x = (new_im.width - text_width) / 2252        y = (new_im.height - text_height) / 2253 254        draw.text((x, y), text, fill=(255, 255, 255))255 256        new_im = np.array(new_im)257        video.write(new_im)258    outputString = ""259    if inital_diff != 10000:260        outputString+=f"Initial difference at frame {inital_diff} at time {inital_diff/fps} seconds"261        outputString+=f"Final difference at frame {final_diff} at time {final_diff/fps} seconds"262    video.release()263    if(outputString==""):264        outputString= "Not tampering are detected"265    return ("video.mp4",outputString)