File size: 7,775 Bytes
127e91a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
"""
Image Forensics for Scientific Figure Manipulation Detection
Specialized for western blots, gel electrophoresis, microscopy images.
"""

import torch
import torch.nn as nn
import numpy as np
from PIL import Image
import cv2


class ForensicFilterBank(nn.Module):
    """Bank of forensic filters for detecting image manipulation in scientific figures."""
    
    def __init__(self):
        super().__init__()
        # SRM (Spatial Rich Model) filters for noise residuals
        self.srm_filters = nn.Conv2d(1, 9, kernel_size=5, padding=2, bias=False)
        srm_kernels = torch.tensor([
            [[0, 0, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, -1, 0, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 1, -1, 0, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 0, -1, 1, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 1, 0], [0, 0, -1, 0, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 0, -1, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 1, -2, 1, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, -2, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 0, 1, 0], [0, 0, -2, 0, 0], [0, 1, 0, 0, 0], [0, 0, 0, 0, 0]],
            [[0, 0, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, -2, 0, 0], [0, 0, 1, 0, 0], [0, 0, 0, 0, 0]]
        ], dtype=torch.float32).unsqueeze(1)
        self.srm_filters.weight = nn.Parameter(srm_kernels)
        for param in self.srm_filters.parameters():
            param.requires_grad = False
        
        # Bayar filter for boundary artifacts
        self.bayar_filter = nn.Conv2d(1, 3, kernel_size=5, padding=2, bias=False)
        bayar_kernel = torch.zeros(3, 1, 5, 5)
        bayar_kernel[0, 0, 2, 2] = -1
        bayar_kernel[0, 0, 2, 1] = 1
        bayar_kernel[1, 0, 2, 2] = -1
        bayar_kernel[1, 0, 1, 2] = 1
        bayar_kernel[2, 0, 2, 2] = -1
        bayar_kernel[2, 0, 2, 3] = 1
        self.bayar_filter.weight = nn.Parameter(bayar_kernel)
        for param in self.bayar_filter.parameters():
            param.requires_grad = False
        
        # Error Level Analysis (ELA) simulation
        self.ela_conv = nn.Sequential(
            nn.Conv2d(3, 16, 3, padding=1), nn.ReLU(),
            nn.Conv2d(16, 16, 3, padding=1), nn.ReLU()
        )
    
    def forward(self, image):
        gray = 0.299 * image[:, 0:1] + 0.587 * image[:, 1:2] + 0.114 * image[:, 2:3]
        srm_out = self.srm_filters(gray)
        bayar_out = self.bayar_filter(gray)
        ela_out = self.ela_conv(image)
        target_size = image.shape[2:]
        srm_out = torch.nn.functional.interpolate(srm_out, size=target_size, mode='bilinear')
        bayar_out = torch.nn.functional.interpolate(bayar_out, size=target_size, mode='bilinear')
        forensic = torch.cat([srm_out, bayar_out, ela_out], dim=1)
        return forensic


class WesternBlotAnalyzer:
    """Specialized analyzer for western blot image manipulation detection."""
    
    @staticmethod
    def detect_band_duplication(image, threshold=0.95):
        if isinstance(image, str):
            img = cv2.imread(image, cv2.IMREAD_GRAYSCALE)
        else:
            img = np.array(image.convert('L'))
        _, binary = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
        num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(binary, connectivity=8)
        bands = []
        for i in range(1, num_labels):
            x, y, w, h, area = stats[i]
            if area > 50:
                band = img[y:y+h, x:x+w]
                bands.append((i, band, (x, y, w, h)))
        duplicates = []
        for i in range(len(bands)):
            for j in range(i+1, len(bands)):
                band1 = bands[i][1]
                band2 = bands[j][1]
                h = max(band1.shape[0], band2.shape[0])
                w = max(band1.shape[1], band2.shape[1])
                b1 = cv2.resize(band1, (w, h))
                b2 = cv2.resize(band2, (w, h))
                similarity = np.corrcoef(b1.flatten(), b2.flatten())[0, 1]
                if similarity > threshold:
                    duplicates.append((bands[i][0], bands[j][0], similarity))
        return duplicates
    
    @staticmethod
    def detect_splicing_artifacts(image):
        if isinstance(image, str):
            img = cv2.imread(image)
        else:
            img = np.array(image)
        gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) if len(img.shape) == 3 else img
        edges = cv2.Canny(gray, 50, 150)
        lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=100, minLineLength=100, maxLineGap=10)
        artifacts = []
        if lines is not None:
            for line in lines:
                x1, y1, x2, y2 = line[0]
                if abs(x2 - x1) > abs(y2 - y1):
                    y = (y1 + y2) // 2
                    region_above = gray[max(0, y-5):y, min(x1,x2):max(x1,x2)]
                    region_below = gray[y:min(gray.shape[0], y+5), min(x1,x2):max(x1,x2)]
                    if region_above.size > 0 and region_below.size > 0:
                        noise_above = np.std(region_above.astype(float))
                        noise_below = np.std(region_below.astype(float))
                        if abs(noise_above - noise_below) > 5:
                            artifacts.append({'line': (x1, y1, x2, y2), 'noise_diff': abs(noise_above - noise_below)})
        return artifacts
    
    @staticmethod
    def compute_manipulation_score(image):
        score = 0.0
        reasons = []
        duplicates = WesternBlotAnalyzer.detect_band_duplication(image)
        if duplicates:
            score += min(0.4, len(duplicates) * 0.1)
            reasons.append(f"Band duplication detected: {len(duplicates)} pairs")
        artifacts = WesternBlotAnalyzer.detect_splicing_artifacts(image)
        if artifacts:
            score += min(0.3, len(artifacts) * 0.05)
            reasons.append(f"Splicing artifacts: {len(artifacts)} boundaries")
        if isinstance(image, str):
            img = cv2.imread(image, cv2.IMREAD_GRAYSCALE)
        else:
            img = np.array(image.convert('L'))
        h, w = img.shape
        sub_size = min(h, w) // 4
        if sub_size > 20:
            hashes = []
            for i in range(0, h - sub_size, sub_size // 2):
                for j in range(0, w - sub_size, sub_size // 2):
                    sub = img[i:i+sub_size, j:j+sub_size]
                    hashes.append((i, j, np.mean(sub)))
            for i in range(len(hashes)):
                for j in range(i+1, len(hashes)):
                    if abs(hashes[i][2] - hashes[j][2]) < 1.0:
                        score += 0.01
        score = min(score, 1.0)
        return score, reasons


class ScientificImageForensics(nn.Module):
    """Complete forensics pipeline for scientific images."""
    
    def __init__(self):
        super().__init__()
        self.filter_bank = ForensicFilterBank()
        self.manipulation_cnn = nn.Sequential(
            nn.Conv2d(28, 32, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
            nn.Conv2d(32, 64, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
            nn.Conv2d(64, 128, 3, padding=1), nn.ReLU(),
            nn.AdaptiveAvgPool2d((4, 4)), nn.Flatten(),
            nn.Linear(128 * 4 * 4, 256), nn.ReLU(), nn.Dropout(0.3),
            nn.Linear(256, 1), nn.Sigmoid()
        )
    
    def forward(self, image):
        forensic_features = self.filter_bank(image)
        manipulation_prob = self.manipulation_cnn(forensic_features)
        return manipulation_prob, forensic_features