Update app.py

app.py

@@ -38,6 +38,10 @@ for key in models:
         models[key]["processor"] = None
         models[key]["model"] = None
 
+
+## 2. 模型输出处理
+
+python
 def process_model_output(model_info, outputs, probabilities):
     """处理不同模型的输出，统一返回AI生成概率"""
     model_name = model_info["name"].lower()
@@ -56,16 +60,16 @@ def process_model_output(model_info, outputs, probabilities):
             if "human" in label_lower or "real" in label_lower:
                 human_label_idx = idx
         
-        # 修改后的标签解释逻辑 - 反转解释
+        # 修正后的标签解释逻辑
         if human_label_idx is not None:
-            # 将human标签视为AI生成的指标
-            ai_probability = float(probabilities[0][human_label_idx].item())
+            # 如果预测为human，则AI概率应该低
+            ai_probability = 1 - float(probabilities[0][human_label_idx].item())
         elif ai_label_idx is not None:
-            # 如果有AI标签
+            # 如果预测为AI，则AI概率应该高
             ai_probability = float(probabilities[0][ai_label_idx].item())
         else:
-            # 默认使用索引1作为AI标签
-            ai_probability = float(probabilities[0][1].item())
+            # 默认情况
+            ai_probability = 0.5
     
     elif "resnet" in model_name:
         # 通用图像分类模型，使用简单启发式方法
@@ -108,6 +112,10 @@ def process_model_output(model_info, outputs, probabilities):
     
     return ai_probability
 
+
+## 3. 图像特征分析
+
+python
 def analyze_image_features(image):
     """分析图像特征"""
     # 转换为OpenCV格式
@@ -203,6 +211,56 @@ def analyze_image_features(image):
     
     return features
 
+
+## 4. AI特征检查
+
+python
+def check_ai_specific_features(image_features):
+    """检查AI生成图像的典型特征"""
+    ai_score = 0
+    ai_signs = []
+    
+    # 检查对称性 - AI生成图像通常对称性高
+    if "horizontal_symmetry" in image_features and "vertical_symmetry" in image_features:
+        avg_symmetry = (image_features["horizontal_symmetry"] + image_features["vertical_symmetry"]) / 2
+        if avg_symmetry > 0.7:
+            ai_score += 0.3
+            ai_signs.append("图像对称性异常高")
+    
+    # 检查纹理相关性 - AI生成图像通常纹理相关性高
+    if "texture_correlation" in image_features and image_features["texture_correlation"] > 0.9:
+        ai_score += 0.2
+        ai_signs.append("纹理相关性异常高")
+    
+    # 检查边缘与噪声的关系 - AI生成图像通常边缘清晰但噪声不自然
+    if "edge_density" in image_features and "noise_level" in image_features:
+        edge_noise_ratio = image_features["edge_density"] / max(image_features["noise_level"], 0.001)
+        if edge_noise_ratio < 0.01:
+            ai_score += 0.2
+            ai_signs.append("边缘与噪声分布不自然")
+    
+    # 检查颜色平滑度 - AI生成图像通常颜色过渡更平滑
+    if "color_std" in image_features and image_features["color_std"] < 10:
+        ai_score += 0.2
+        ai_signs.append("颜色过渡异常平滑")
+    
+    # 检查纹理能量 - AI生成图像通常纹理能量分布不自然
+    if "texture_energy" in image_features and image_features["texture_energy"] < 0.02:
+        ai_score += 0.2
+        ai_signs.append("纹理能量分布不自然")
+    
+    # 检查频率比例 - AI生成图像通常频率分布不自然
+    if "freq_ratio" in image_features:
+        if image_features["freq_ratio"] < 0.1 or image_features["freq_ratio"] > 2.0:
+            ai_score += 0.2
+            ai_signs.append("频率分布不自然")
+    
+    return min(ai_score, 1.0), ai_signs
+
+
+## 5. PS痕迹检测
+
+python
 def detect_photoshop_signs(image_features):
     """检测图像中的PS痕迹"""
     ps_score = 0
@@ -254,16 +312,13 @@ def detect_photoshop_signs(image_features):
             ps_score += 0.2
             ps_signs.append("频率分布不自然，可能有过度锐化处理")
     
-    # 检查对称性不自然
-    if "horizontal_symmetry" in image_features and "vertical_symmetry" in image_features:
-        avg_symmetry = (image_features["horizontal_symmetry"] + image_features["vertical_symmetry"]) / 2
-        if avg_symmetry > 0.8:
-            ps_score += 0.2
-            ps_signs.append("图像对称性过高，可能经过对称处理")
-    
     return min(ps_score, 1.0), ps_signs
 
-def get_detailed_analysis(ai_probability, ps_score, ps_signs, valid_models_count):
+
+## 6. 结果分析与分类
+
+python
+def get_detailed_analysis(ai_probability, ps_score, ps_signs, ai_signs, valid_models_count):
     """提供更详细的分析结果"""
     
     # 根据有效模型数量调整置信度描述
@@ -275,17 +330,18 @@ def get_detailed_analysis(ai_probability, ps_score, ps_signs, valid_models_count
     elif valid_models_count == 1:
         confidence_prefix = "中等置信度："
     
-    if ai_probability > 0.8:
+    # 调整后的阈值判断
+    if ai_probability > 0.6:  # 降低为0.6
         category = confidence_prefix + "高概率AI生成"
-        description = "图像几乎可以确定是由AI完全生成，几乎没有真人照片的特征。"
-    elif ai_probability > 0.6:
+        description = "图像很可能是由AI完全生成，几乎没有真人照片的特征。"
+    elif ai_probability > 0.4:  # 降低为0.4
         if ps_score > 0.5:
             category = confidence_prefix + "中等概率AI生成，高概率PS修图"
             description = "图像可能是真人照片经过大量后期处理，或是AI生成后经过修饰的图像。"
         else:
             category = confidence_prefix + "中等概率AI生成"
             description = "图像有较多AI生成的特征，但也保留了一些真实照片的特点。"
-    elif ai_probability > 0.4:
+    elif ai_probability > 0.3:  # 降低为0.3
         if ps_score > 0.5:
             category = confidence_prefix + "低概率AI生成，高概率PS修图"
             description = "图像更可能是真人照片经过大量后期处理，PS痕迹明显。"
@@ -312,8 +368,18 @@ def get_detailed_analysis(ai_probability, ps_score, ps_signs, valid_models_count
     else:
         ps_details = "未检测到明显的修图痕迹。"
     
-    return category, description, ps_details
+    # 添加AI特征描述
+    if ai_signs:
+        ai_details = "检测到的AI特征：" + "、".join(ai_signs)
+    else:
+        ai_details = "未检测到明显的AI生成特征。"
+    
+    return category, description, ps_details, ai_details
+
 
+## 7. 主检测函数
+
+python
 def detect_ai_image(image):
     """主检测函数"""
     if image is None:
@@ -365,47 +431,70 @@ def detect_ai_image(image):
     # 分析图像特征
     image_features = analyze_image_features(image)
     
+    # 检查AI特定特征
+    ai_feature_score, ai_signs = check_ai_specific_features(image_features)
+    
+    # 分析PS痕迹
+    ps_score, ps_signs = detect_photoshop_signs(image_features)
+    
     # 应用特征权重调整AI概率
     adjusted_probability = final_ai_probability
     
-    # 低边缘密度通常表示AI生成
-    if image_features["edge_density"] < 0.01:
-        adjusted_probability += 0.15
+    # 如果AI特征分数高，大幅提高AI概率
+    if ai_feature_score > 0.5:
+        adjusted_probability = max(adjusted_probability, 0.7)
+    elif ai_feature_score > 0.3:
+        adjusted_probability = max(adjusted_probability, 0.5)
     
-    # 高纹理均匀性通常表示AI生成
-    if "texture_homogeneity" in image_features and image_features["texture_homogeneity"] > 0.5:
-        adjusted_probability += 0.1
+    # 高对称性是AI生成的强烈指标
+    if "horizontal_symmetry" in image_features and image_features["horizontal_symmetry"] > 0.7:
+        adjusted_probability += 0.15
+    if "vertical_symmetry" in image_features and image_features["vertical_symmetry"] > 0.7:
+        adjusted_probability += 0.15
     
-    # 低噪声水平通常表示AI生成
-    if image_features["noise_level"] < 0.5:
+    # 高纹理相关性通常表示AI生成
+    if "texture_correlation" in image_features and image_features["texture_correlation"] > 0.9:
         adjusted_probability += 0.1
     
-    # 高对称性通常表示AI生成
-    if "horizontal_symmetry" in image_features and image_features["horizontal_symmetry"] > 0.8:
+    # 低边缘密度通常表示AI生成
+    if image_features["edge_density"] < 0.01:
         adjusted_probability += 0.1
     
     # 确保概率在0-1范围内
     adjusted_probability = min(1.0, max(0.0, adjusted_probability))
     
-    # 分析PS痕迹
-    ps_score, ps_signs = detect_photoshop_signs(image_features)
+    # 如果umm-maybe/AI-image-detector模型的预测与其他模型不一致，增加其权重
+    if "model1" in results and "ai_probability" in results["model1"]:
+        ai_detector_prob = results["model1"]["ai_probability"]
+        # 如果专用AI检测器给出的概率与调整后概率差异大，增加其权重
+        if abs(ai_detector_prob - adjusted_probability) > 0.3:
+            adjusted_probability = (adjusted_probability + ai_detector_prob * 2) / 3
     
     # 获取详细分析
-    category, description, ps_details = get_detailed_analysis(adjusted_probability, ps_score, ps_signs, valid_models)
+    category, description, ps_details, ai_details = get_detailed_analysis(
+        adjusted_probability, ps_score, ps_signs, ai_signs, valid_models
+    )
     
     # 构建最终结果
     final_result = {
         "ai_probability": adjusted_probability,
+        "original_ai_probability": final_ai_probability,
         "ps_score": ps_score,
+        "ai_feature_score": ai_feature_score,
         "category": category,
         "description": description,
         "ps_details": ps_details,
+        "ai_details": ai_details,
         "individual_model_results": results,
         "features": image_features
     }
     
     return final_result
 
+
+## 8. Gradio界面
+
+python
 # 创建Gradio界面
 iface = gr.Interface(
     fn=detect_ai_image,