Initial Docker deployment of LSB Steganography Flask app

Dockerfile

@@ -0,0 +1,11 @@
+FROM python:3.10
+
+WORKDIR /app
+
+COPY . /app
+
+RUN pip install --no-cache-dir -r requirements.txt
+
+EXPOSE 7860
+
+CMD ["python", "app.py"]

app.py

@@ -0,0 +1,166 @@
+from flask import Flask, render_template, request, jsonify
+import torch
+import torch.nn as nn
+from PIL import Image
+import numpy as np
+import io
+import pickle
+from scipy.stats import entropy
+
+app = Flask(__name__)
+app.config['MAX_CONTENT_LENGTH'] = 5 * 1024 * 1024
+
+# --------------------
+# Model definitions
+# --------------------
+class BinaryDNN(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.Linear(196614, 384), nn.BatchNorm1d(384), nn.ReLU(), nn.Dropout(0.45),
+            nn.Linear(384, 192), nn.BatchNorm1d(192), nn.ReLU(), nn.Dropout(0.35),
+            nn.Linear(192, 96), nn.BatchNorm1d(96), nn.ReLU(), nn.Dropout(0.25),
+            nn.Linear(96, 2))
+    def forward(self, x): return self.net(x)
+
+class MultiDNN(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.Linear(196614, 384), nn.BatchNorm1d(384), nn.ReLU(), nn.Dropout(0.45),
+            nn.Linear(384, 192), nn.BatchNorm1d(192), nn.ReLU(), nn.Dropout(0.35),
+            nn.Linear(192, 96), nn.BatchNorm1d(96), nn.ReLU(), nn.Dropout(0.25),
+            nn.Linear(96, 4))
+    def forward(self, x): return self.net(x)
+
+# --------------------
+# Load artifacts ONCE
+# --------------------
+with open('final_scaler.pkl', 'rb') as f:
+    scaler = pickle.load(f)
+with open('final_bin_le.pkl', 'rb') as f:
+    binary_le = pickle.load(f)
+with open('final_multi_le.pkl', 'rb') as f:
+    multi_le = pickle.load(f)
+
+device = torch.device('cpu')
+binary_model = BinaryDNN()
+binary_model.load_state_dict(torch.load('final_bin_best.pth', map_location=device))
+binary_model.eval()
+multi_model = MultiDNN()
+multi_model.load_state_dict(torch.load('final_multi_best.pth', map_location=device))
+multi_model.eval()
+
+# --------------------
+# V3 Feature extraction
+# --------------------
+def extract_lsb_feature(img, k=2):
+    mask = (1 << k) - 1
+    return (img & mask).flatten()
+
+def extract_randomness_features(lsb_feat):
+    chunk_size = 512
+    lsb_reshaped = lsb_feat.reshape(-1, chunk_size)
+    ent_values = []
+    for block in lsb_reshaped[::10]:
+        hist = np.bincount(block.astype(np.int32), minlength=4)
+        hist = hist / max(hist.sum(), 1)
+        ent = entropy(hist + 1e-10)
+        ent_values.append(ent)
+    mean_entropy = np.mean(ent_values)
+    sample = lsb_feat[:5000].astype(np.int32)
+    transitions = np.sum(np.abs(np.diff(sample)))
+    transition_rate = transitions / max(len(sample) - 1, 1)
+    hist_full = np.bincount(lsb_feat.astype(np.int32), minlength=4)
+    expected = len(lsb_feat) / 4
+    chi2_stat = np.sum((hist_full - expected) ** 2 / expected) if expected > 0 else 0
+    return np.array([mean_entropy, transition_rate, chi2_stat], dtype=np.float32)
+
+def extract_url_pattern_features(lsb_feat):
+    chunk_size = 4
+    num_chunks = len(lsb_feat) // chunk_size
+    lsb_trimmed = lsb_feat[:num_chunks * chunk_size].astype(np.int32)
+    byte_values = []
+    for i in range(0, len(lsb_trimmed), chunk_size):
+        val = (lsb_trimmed[i] << 6) | (lsb_trimmed[i+1] << 4) | (lsb_trimmed[i+2] << 2) | lsb_trimmed[i+3]
+        byte_values.append(val)
+    byte_values = np.array(byte_values)
+    url_char_count = np.sum((byte_values >= 10) & (byte_values <= 15))
+    html_char_count = np.sum((byte_values >= 0) & (byte_values <= 5))
+    unique_ratio = len(np.unique(byte_values)) / max(len(byte_values), 1)
+    return np.array([
+        url_char_count / max(len(byte_values), 1),
+        html_char_count / max(len(byte_values), 1),
+        unique_ratio
+    ], dtype=np.float32)
+
+def extract_features(image, k=2):
+    image = image.resize((256, 256), Image.LANCZOS)
+    if image.mode != 'RGB':
+        image = image.convert('RGB')
+    img_array = np.array(image)
+    lsb_feat = extract_lsb_feature(img_array, k=k)
+    rand_feat = extract_randomness_features(lsb_feat)
+    url_feat = extract_url_pattern_features(lsb_feat)
+    combined = np.concatenate([lsb_feat.astype(np.float32), rand_feat, url_feat])
+    return combined
+
+# --------------------
+# Flask endpoints
+# --------------------
+@app.route('/')
+def index():
+    return render_template('index.html')
+
+@app.route('/predict', methods=['POST'])
+def predict():
+    try:
+        file = request.files['file']
+        image = Image.open(io.BytesIO(file.read()))
+        features = extract_features(image)
+        features_normalized = scaler.transform(features.reshape(1, -1)).astype(np.float32)
+        features_tensor = torch.FloatTensor(features_normalized)
+
+        with torch.no_grad():
+            binary_output = binary_model(features_tensor)
+            binary_probs = torch.softmax(binary_output, dim=1).cpu().numpy()[0]
+            binary_pred_idx = int(np.argmax(binary_probs))
+            binary_class = binary_le.classes_[binary_pred_idx]
+        
+        if binary_class == 'stego':
+            with torch.no_grad():
+                multi_output = multi_model(features_tensor)
+                multi_probs = torch.softmax(multi_output, dim=1).cpu().numpy()[0]
+                multi_pred_idx = int(np.argmax(multi_probs))
+                payload_type = multi_le.classes_[multi_pred_idx]
+        else:
+            payload_type = 'clean'
+            multi_probs = np.zeros(len(multi_le.classes_))
+            multi_probs[multi_le.transform(['clean'])[0]] = 1.0
+
+        result = {
+            'binary_prediction': {
+                'class': binary_class,
+                'confidence': float(binary_probs[binary_pred_idx]),
+                'probabilities': {c: float(binary_probs[binary_le.transform([c])[0]]) for c in binary_le.classes_}
+            },
+            'multiclass_prediction': {
+                'payload_type': payload_type,
+                'probabilities': {c: float(multi_probs[i]) for i, c in enumerate(multi_le.classes_)}
+            }
+        }
+        return jsonify(result)
+    except Exception as e:
+        import traceback
+        print('Error:', str(e))
+        traceback.print_exc()
+        return jsonify({'error': str(e)}), 500
+
+@app.errorhandler(413)
+def request_entity_too_large(error):
+    return jsonify({'error': 'File size exceeds 5MB'}), 413
+
+
+if __name__ == "__main__":
+    port = int(os.environ.get("PORT", 7860))
+    app.run(debug=True, host="0.0.0.0", port=port)

final_bin_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:937140babfa952b004688c3bb98665434afd2b901658e5e8ff4de0438c107d90
+size 302389710

final_bin_le.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90b3bd41467e805a270b7d2b1ca04eb1ea9d31670d16bf5c4022f81f5c3c6d9f
+size 276

final_multi_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d10330083813ade2f03ff8365b24abdc732f5728f1470372390863f748073ee
+size 302390532

final_multi_le.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b16da8d351473e04c4cf7ebcac502f4f7410e7832ec44db0f2e2f1c8d6ddf93b
+size 316

final_scaler.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a115a9cf586486270825e41369e3d3e6f2367f6f49c16f1d583d6ac626015a2
+size 4719234

requirements.txt

@@ -0,0 +1,6 @@
+flask
+torch
+numpy
+scipy
+pillow
+scikit-learn

templates/index.html

@@ -0,0 +1,223 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <title>Multi-Payload LSB Steganalysis Demo</title>
+  <meta name="viewport" content="width=device-width, initial-scale=1">
+  <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.0/dist/css/bootstrap.min.css" rel="stylesheet">
+  <link href="https://fonts.googleapis.com/css?family=Inter:600,700&amp;display=swap" rel="stylesheet">
+  <style>
+    body {
+      min-height: 100vh;
+      background: linear-gradient(120deg,#090e1a 0%,#1e293b 75%,#64748b 100%);
+      font-family: 'Inter', Arial, sans-serif;
+      margin:0;padding:0;
+      overflow-x: hidden;
+    }
+    .main-card {
+      background: rgba(255,255,255,0.92);
+      border-radius: 2rem;
+      box-shadow: 0 12px 40px rgba(20,32,64,0.21);
+      margin: 2.2rem auto;
+      max-width: 790px;
+      width: 98vw;
+      padding: 2.7rem 2.2rem 2.7rem 2.2rem;
+      transition: box-shadow 0.18s;
+      position: relative;
+      backdrop-filter: blur(2px);
+    }
+    .header-title {
+      font-size: 2.4rem;
+      font-weight: 900;
+      color: #2563eb;
+      text-align: center;
+      letter-spacing: .016em;
+      margin-bottom: 0.18em;
+      text-shadow: 0 3px 12px #64748b26;
+    }
+    .header-desc {
+      text-align: center;
+      color: #334155;
+      font-size: 1.22rem;
+      font-weight: 600;
+      margin-bottom: 1.5em;
+      letter-spacing: 0.07em;
+    }
+    /* Glass info panel */
+    .info-glass {
+      background: rgba(229,240,255,0.82);
+      border-radius: 1.1rem;
+      box-shadow: 0 2px 18px #0797c522, 0 0.5px 3.5px #ffd70033;
+      padding: 1.3em 1.3em 0.9em 1.3em;
+      margin-bottom: 1.45em;
+    }
+    .info-title { color: #2563eb; font-size: 1.12em; font-weight: 700; margin-bottom: 0.7em;}
+    .info-snip { color: #23304f; font-size: 0.97em; margin-bottom: 0.3em;}
+    .info-payloads td, .info-payloads th {padding: 0.26em .8em;}
+    .info-collapsible { cursor: pointer; font-size: 1.07rem; color: #334155; background:none;border:none;font-weight:700;}
+    .info-collapsible:after {content: "▼";font-size:.9em;margin-left:.5em;}
+    .info-collapsible.collapsed:after {content:"►";}
+    .info-content {display:block;transition: max-height .3s cubic-bezier(.32,1.8,.75,1);}
+    .info-content.collapsed {display:none;}
+    .dropzone { border: 3px dashed #2563eb; border-radius: 1.2rem; text-align: center; background: #f5faff; padding: 2.3rem 1.1rem; cursor: pointer; transition: border-color 0.22s, background 0.2s, box-shadow .21s;}
+    .dropzone:hover { background: #dbeafe; border-color: #3b82f6; box-shadow: 0 3px 26px #2563eb33; }
+    .preview-img { max-width: 250px; border-radius: 1rem; margin-top: 1rem; box-shadow: 0 0 16px #3b82f655; background: #fff;}
+    .result-badge { font-size: 1.2em; font-weight: 800; padding: 0.45em 1.6em; border-radius: 2em; letter-spacing: 0.03em; display:inline-block; margin-bottom:0.3em; box-shadow: 0 3px 10px #0002;}
+    .prob-bar-wrap { display: flex; align-items: center; gap:0.45em; margin-bottom: 0.6em;}
+    .prob-bar-label { width:80px; font-weight:700; font-size:1.09em; color:#222b3a; letter-spacing:.01em; text-transform:uppercase; text-align:right;}
+    .prob-bar { flex: 1; height: 29px; border-radius: 1em; background: #e0e7ef; overflow: hidden; position:relative; margin-right:6px;}
+    .prob-bar-inner { height:100%;display:flex;align-items:center; font-weight:700;color:#fff;font-size:1.17em;padding-left:1em; border-radius:1em;transition:width 0.55s cubic-bezier(.32,1.2,.69,1);}
+    .prob-bar-value { min-width:52px;text-align:right;font-weight:600;font-size:1.05em;margin-left:4px; color:#1e293b;}
+    .icon-big { font-size: 2.2em; vertical-align: sub; margin-right: 0.23em;}
+    .icon-clean { color: #1dd8a2;}
+    .icon-stego { color: #ef4444;}
+    /* Subtle watermark background icon */
+    .bg-watermark {
+      position: absolute; bottom: 14px; right: 18px; z-index:0; opacity:.12; font-size:5.5em; pointer-events:none;
+    }
+    @media(max-width: 900px){ .main-card{ max-width:99vw; padding:1.3rem}}
+    @media(max-width: 520px){ .prob-bar-label{width:54px;font-size:0.99em}}
+  </style>
+</head>
+<body>
+<div class="container-fluid">
+  <div class="main-card">
+    <!-- Watermark icon -->
+    <div class="bg-watermark"><span class="icon-big icon-stego">🔒</span></div>
+    <!-- Header section -->
+    <div class="header-title">Multi-Payload LSB Steganalysis Demo</div>
+    <div class="header-desc">
+      Analyze PNG images for hidden data using advanced<br>Least Significant Bit detection and AI.<br>
+      Upload an image and instantly detect steganographic payloads!
+    </div>
+    <!-- Info section -->
+    <div class="info-glass mb-4">
+      <button class="info-collapsible" onclick="toggleInfo()">About & Knowledge  </button>
+      <div id="infoPanel" class="info-content collapsed">
+        <div class="info-title">What is Steganography?</div>
+        <div class="info-snip">Steganography is the science of hiding information within digital media so it is invisible to regular inspection. Images, audio, and video files are common carriers.</div>
+        <div class="info-title mt-2">What are LSBs and Payloads?</div>
+        <div class="info-snip">LSB (Least Significant Bit) steganography modifies the lowest-value bits in an image (usually the last 1 or 2 of each pixel) to embed secret data called the payload.</div>
+        <div class="info-title mt-2">What Payload Types Are Detected?</div>
+        <table class="info-payloads mb-1 small">
+          <tr><th>Payload</th><th>Meaning</th></tr>
+          <tr><td>HTML</td><td>Hidden HTML code</td></tr>
+          <tr><td>JS</td><td>Hidden JavaScript code</td></tr>
+          <tr><td>PS</td><td>Hidden Powershell script</td></tr>
+          <tr><td>Clean</td><td>No payload / genuine image</td></tr>
+        </table>
+        <div class="info-title mt-2">How does the model decide?</div>
+        <div class="info-snip mb-2">The neural network analyzes subtle changes in the 2 least significant bits of every image pixel, plus additional statistics, to predict if a payload is present and its type. Confidence is shown for each class.</div>
+      </div>
+    </div>
+    <!-- Upload & results UI -->
+    <div class="dropzone" id="dropzone">
+      <span class="fs-3">🖼️</span>
+      <div class="mt-2 mb-1 fw-semibold">Upload PNG image or drag & drop</div>
+      <div class="text-muted small mb-0">Max 5MB &nbsp;|&nbsp; Secure PyTorch Model</div>
+      <input type="file" id="fileInput" accept="image/png" style="display:none;">
+    </div>
+    <div id="preview" style="display:none;" class="text-center mt-3">
+      <img id="previewImg" class="preview-img mb-2" alt="Preview"/>
+      <div id="filename" class="small text-muted mt-1"></div>
+    </div>
+    <button id="analyzeBtn" class="btn btn-primary w-100 mt-3 fw-bold" disabled>Analyze Image</button>
+    <div id="loading" style="display:none;" class="text-center mt-3">
+      <div class="spinner-border text-primary" style="width:2.1rem;height:2.1rem;"></div>
+      <p class="mt-2">Analyzing image...</p>
+    </div>
+    <div id="results" style="display:none;" class="mt-4"></div>
+  </div>
+</div>
+<script>
+let selectedFile = null;
+const dropzone = document.getElementById('dropzone');
+const fileInput = document.getElementById('fileInput');
+const preview = document.getElementById('preview');
+const previewImg = document.getElementById('previewImg');
+const filenameDiv = document.getElementById('filename');
+const analyzeBtn = document.getElementById('analyzeBtn');
+const loading = document.getElementById('loading');
+const results = document.getElementById('results');
+const infoPanel = document.getElementById('infoPanel');
+
+dropzone.onclick = () => fileInput.click();
+dropzone.ondragover = (e) => { e.preventDefault(); dropzone.style.background = "#dbeafe"; };
+dropzone.ondragleave = () => dropzone.style.background = "#f5faff";
+dropzone.ondrop = (e) => {
+    e.preventDefault(); dropzone.style.background = "#dbeafe";
+    if (e.dataTransfer.files.length) handleFile(e.dataTransfer.files[0]);
+};
+fileInput.onchange = (e) => { if (e.target.files.length) handleFile(e.target.files[0]); };
+function handleFile(file) {
+    if (!file.name.toLowerCase().endsWith('.png')) { alert('Please select a PNG image.'); return; }
+    selectedFile = file; analyzeBtn.disabled = false;
+    preview.style.display = 'block';
+    filenameDiv.textContent = `${file.name} (${(file.size/1024).toFixed(1)} KB)`;
+    const reader = new FileReader();
+    reader.onload = e => { previewImg.src = e.target.result; }; reader.readAsDataURL(file);
+    results.style.display = 'none';
+}
+analyzeBtn.onclick = async () => {
+    if (!selectedFile) return;
+    loading.style.display = 'block'; results.style.display = 'none'; analyzeBtn.disabled = true;
+    const formData = new FormData(); formData.append('file', selectedFile);
+    try {
+        const resp = await fetch('/predict', { method:'POST', body:formData });
+        const data = await resp.json();
+        if (resp.ok) displayResults(data); else alert(data.error || 'Prediction failed');
+    } catch (e) { alert('Network error: ' + e.message); }
+    loading.style.display = 'none'; analyzeBtn.disabled = false;
+};
+// Collapsible info
+function toggleInfo(){
+    infoPanel.classList.toggle('collapsed');
+    document.querySelector('.info-collapsible').classList.toggle('collapsed');
+}
+// Custom icons for clean/stego
+const CLEAN_EMOJI = '<span class="icon-big icon-clean">🔓</span>';
+const STEGO_EMOJI = '<span class="icon-big icon-stego">🔒</span>';
+function displayResults(data) {
+    let html = '';
+    const b = data.binary_prediction, m = data.multiclass_prediction;
+    html += `<div class="mb-2 text-center">
+        <span class="result-badge" 
+            style="background:${b.class==='clean'?'#14b8a6':'#ef4444'};color:#fff">
+            ${b.class==='clean' ? CLEAN_EMOJI+'CLEAN' : STEGO_EMOJI+'STEGO'}
+        </span>
+        <div class="small mt-1 mb-2" style="color:#34425a;">Binary prediction (Clean vs. Stego)</div>
+        <div class="prob-bar-wrap">
+            <span class="prob-bar-label">CLEAN</span>
+            <div class="prob-bar"><div class="prob-bar-inner" style="width:${(b.probabilities.clean*100).toFixed(1)}%;background:#10b981;"></div></div>
+            <span class="prob-bar-value">${(b.probabilities.clean*100).toFixed(1)}%</span>
+        </div>
+        <div class="prob-bar-wrap">
+            <span class="prob-bar-label">STEGO</span>
+            <div class="prob-bar"><div class="prob-bar-inner" style="width:${(b.probabilities.stego*100).toFixed(1)}%;background:#ef4444;"></div></div>
+            <span class="prob-bar-value">${(b.probabilities.stego*100).toFixed(1)}%</span>
+        </div>
+    </div>`;
+    if (b.class === 'stego') {
+        html += `<div class="mb-2 text-center">
+            <span class="result-badge" style="background:#3b82f6;color:#fff">Payload: ${m.payload_type.toUpperCase()}</span>
+            <div class="small mb-2" style="color:#2a3140;">Multi-class payload classification</div>`;
+        const colors = {html:'#6366f1',js:'#f97316',ps:'#14b8a6',clean:'#06b6d4'};
+        Object.entries(m.probabilities)
+            .sort((a,b)=>b[1]-a[1])
+            .forEach(([k,v]) => {
+                let color = colors[k] || '#818cf8';
+                html += `<div class="prob-bar-wrap">
+                    <span class="prob-bar-label">${k.toUpperCase()}</span>
+                    <div class="prob-bar">
+                        <div class="prob-bar-inner" style="width:${(v*100).toFixed(1)}%;background:${color};"></div>
+                    </div>
+                    <span class="prob-bar-value">${(v*100).toFixed(1)}%</span>
+                </div>`;
+            });
+        html += `</div>`;
+    }
+    results.innerHTML = html; results.style.display = 'block';
+}
+</script>
+</body>
+</html>