Update app.py

app.py

@@ -1,90 +1,97 @@
 import gradio as gr
-import torch
-import torchvision.transforms as T
-from PIL import Image
 import numpy as np
+from PIL import Image
+import cv2
 
 # -----------------------------
-# NOTE FOR USER:
-# This is a PROTOTYPE detector.
-# It uses simple CNN-based feature heuristics + frequency analysis.
-# It is NOT a forensic-grade deepfake detector.
+# STEP 1: FACE DETECTION + ROUTING (VERSION 2 - FOUNDATION)
+# -----------------------------
+# This version does NOT yet classify deepfakes.
+# It ONLY decides:
+# 1) Does the image contain a human face?
+# 2) Route it to the correct analysis pipeline
 # -----------------------------
 
-# Image preprocessing
-transform = T.Compose([
-    T.Resize((224, 224)),
-    T.ToTensor(),
-    T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
-])
-
-# Simple frequency analysis (FFT-based heuristic)
-def frequency_artifacts(image: Image.Image):
-    gray = image.convert("L")
-    arr = np.array(gray)
-    fft = np.fft.fft2(arr)
-    fft_shift = np.fft.fftshift(fft)
-    magnitude = np.log(np.abs(fft_shift) + 1)
-    score = np.mean(magnitude)
-    return score
-
-# Main prediction function
-def detect_authenticity(image):
+# Load OpenCV Haar Cascade (lightweight & HF-friendly)
+face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
+
+
+def detect_faces(image_np):
+    """
+    Detect faces using OpenCV Haar Cascade
+    Returns number of faces detected
+    """
+    gray = cv2.cvtColor(image_np, cv2.COLOR_RGB2GRAY)
+    faces = face_cascade.detectMultiScale(
+        gray,
+        scaleFactor=1.1,
+        minNeighbors=5,
+        minSize=(60, 60)
+    )
+    return faces
+
+
+def route_image(image):
     if image is None:
-        return "No image provided", 0.0
+        return "No image provided", "N/A"
 
-    pil_image = Image.fromarray(image)
-    freq_score = frequency_artifacts(pil_image)
+    image_np = np.array(image)
+    faces = detect_faces(image_np)
 
-    # Heuristic thresholds (tunable)
-    if freq_score > 5.2:
-        label = "⚠️ Likely Manipulated / Deepfake"
-        confidence = min((freq_score - 5) * 20, 95)
+    # ROUTING LOGIC
+    if len(faces) > 0:
+        route = "🧑 Face Detected"
+        explanation = (
+            "This image contains one or more human faces.\n"
+            "→ It will be analyzed using **deepfake face-detection models**\n"
+            "(XceptionNet / EfficientNet in next step)."
+        )
     else:
-        label = "✅ Likely Authentic"
-        confidence = min((6 - freq_score) * 20, 95)
+        route = "🖼️ No Face Detected"
+        explanation = (
+            "No human face detected in this image.\n"
+            "→ It will be analyzed using **generic AI-image detection models**\n"
+            "(GAN / diffusion detection)."
+        )
+
+    return route, explanation
 
-    return label, round(confidence, 2)
 
-# Gradio UI
+# -----------------------------
+# GRADIO UI
+# -----------------------------
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
-    # 🕵️ Deepfake & Image Authenticity Detector
-    Upload an image to analyze whether it appears **authentic or manipulated**.
+    # 🧭 Image Analysis Router (Version 2 – Step 1)
 
-    **Use cases:**
-    - Media verification
-    - Social media fact-checking
-    - Educational demos
-    - Digital forensics awareness
+    This step determines **how the image should be analyzed**:
+    - 👤 Face present → Deepfake detection pipeline
+    - 🖼️ No face → Generic AI-image detection pipeline
 
-    ⚠️ *This tool uses heuristic signals and is for educational purposes.*
+    *(No deepfake classification is performed yet.)*
     """)
 
     with gr.Row():
         with gr.Column():
-            image_input = gr.Image(label="Upload Image", type="numpy")
+            image_input = gr.Image(type="pil", label="Upload Any Image")
             analyze_btn = gr.Button("Analyze Image 🔍")
         with gr.Column():
-            result_label = gr.Textbox(label="Prediction")
-            confidence = gr.Slider(0, 100, label="Confidence (%)")
+            route_output = gr.Textbox(label="Routing Decision")
+            explanation_output = gr.Textbox(label="Explanation", lines=5)
 
     analyze_btn.click(
-        fn=detect_authenticity,
+        fn=route_image,
         inputs=image_input,
-        outputs=[result_label, confidence]
+        outputs=[route_output, explanation_output]
     )
 
     gr.Markdown("""
-    ### ℹ️ How it works (Simplified)
-    - Analyzes **frequency artifacts** commonly introduced by GANs & image manipulation
-    - Detects unnatural pixel distributions
-
-    🔧 *Future upgrades you can add:*
-    - XceptionNet / EfficientNet deepfake models
-    - Face-only region detection
-    - Video deepfake detection
-    - Explainable heatmaps
+    ### ✅ Why this step matters
+    - Prevents misuse of face-only deepfake models
+    - Reduces false positives
+    - Makes the system work for **ANY image**
+
+    🔜 **Next step:** Integrate XceptionNet / EfficientNet classifiers.
     """)
 
 # Launch