sameersyed/Defence_FrameWork

1import numpy as np2from PIL import Image, ImageFilter3import io4import base645 6class DefenseSimulator:7    """Simulate defense mechanisms and show before/after results"""8    9    def __init__(self):10        self.defenses = {11            'jpeg_compression': self._jpeg_compression,12            'gaussian_blur': self._gaussian_blur,13            'bit_depth_reduction': self._bit_depth_reduction14        }15    16    def apply_defense(self, image_data, defense_type='jpeg_compression'):17        """Apply defense and return comparison"""18        try:19            # Decode image20            if ',' in image_data:21                image_data = image_data.split(',')[1]22            image_bytes = base64.b64decode(image_data)23            image = Image.open(io.BytesIO(image_bytes)).convert('RGB')24            25            # Apply defense26            defended_image = self.defenses[defense_type](image)27            28            # Encode defended image29            buffered = io.BytesIO()30            defended_image.save(buffered, format="PNG")31            defended_base64 = base64.b64encode(buffered.getvalue()).decode()32            33            # Calculate metrics34            original_array = np.array(image)35            defended_array = np.array(defended_image)36            37            mse = np.mean((original_array - defended_array) ** 2)38            psnr = 10 * np.log10(255**2 / (mse + 1e-10))39            40            return {41                'success': True,42                'defense_type': defense_type,43                'original_image': f"data:image/png;base64,{image_data}",44                'defended_image': f"data:image/png;base64,{defended_base64}",45                'metrics': {46                    'mse': float(mse),47                    'psnr': float(psnr)48                },49                'effectiveness': self._estimate_effectiveness(defense_type)50            }51        52        except Exception as e:53            return {'success': False, 'error': str(e)}54    55    def _jpeg_compression(self, image, quality=75):56        """JPEG compression defense"""57        buffered = io.BytesIO()58        image.save(buffered, format="JPEG", quality=quality)59        buffered.seek(0)60        return Image.open(buffered).convert('RGB')61    62    def _gaussian_blur(self, image, radius=1):63        """Gaussian blur defense"""64        return image.filter(ImageFilter.GaussianBlur(radius=radius))65    66    def _bit_depth_reduction(self, image, bits=6):67        """Bit depth reduction defense"""68        img_array = np.array(image)69        factor = 2 ** (8 - bits)70        reduced = (img_array // factor) * factor71        return Image.fromarray(reduced.astype(np.uint8))72    73    def _estimate_effectiveness(self, defense_type):74        """Estimate defense effectiveness"""75        effectiveness = {76            'jpeg_compression': {'accuracy_retention': 92, 'attack_mitigation': 65},77            'gaussian_blur': {'accuracy_retention': 88, 'attack_mitigation': 70},78            'bit_depth_reduction': {'accuracy_retention': 85, 'attack_mitigation': 60}79        }80        return effectiveness.get(defense_type, {'accuracy_retention': 90, 'attack_mitigation': 60})81    82    def generate_report(self, result):83        """Generate defense comparison report"""84        if not result['success']:85            return f"❌ Defense simulation failed: {result['error']}"86        87        report = []88        report.append("🛡️ Defense Mechanism Simulation\n\n")89        report.append("─" * 60 + "\n\n")90        91        defense_names = {92            'jpeg_compression': 'JPEG Compression (Quality=75)',93            'gaussian_blur': 'Gaussian Blur (Radius=1)',94            'bit_depth_reduction': 'Bit Depth Reduction (6-bit)'95        }96        97        report.append(f"🔧 Applied Defense: {defense_names.get(result['defense_type'], result['defense_type'])}\n\n")98        99        report.append("📊 Image Quality Metrics:\n")100        report.append(f"• MSE (Mean Squared Error): {result['metrics']['mse']:.2f}\n")101        report.append(f"• PSNR (Peak Signal-to-Noise Ratio): {result['metrics']['psnr']:.2f} dB\n\n")102        103        eff = result['effectiveness']104        report.append("📈 Defense Effectiveness:\n")105        report.append(f"• Accuracy Retention: {eff['accuracy_retention']}%\n")106        report.append(f"• Attack Mitigation: {eff['attack_mitigation']}%\n\n")107        108        report.append("─" * 60 + "\n\n")109        report.append("💡 Analysis:\n")110        111        if eff['accuracy_retention'] > 90:112            report.append("• ✅ High accuracy retention - minimal impact on clean samples\n")113        else:114            report.append("• ⚠️ Moderate accuracy retention - some clean accuracy loss\n")115        116        if eff['attack_mitigation'] > 65:117            report.append("• ✅ Strong attack mitigation - effective against adversarial examples\n")118        else:119            report.append("• ⚠️ Moderate attack mitigation - partial protection only\n")120        121        report.append("\n🎯 Recommendation:\n")122        report.append("• Combine multiple defenses for better robustness\n")123        report.append("• Use adversarial training alongside preprocessing\n")124        report.append("• Monitor accuracy-robustness trade-off in production\n")125        126        return "".join(report)127