import os import numpy as np import cv2 import gradio as gr from tensorflow.keras.applications import ResNet50 from tensorflow.keras.applications.resnet50 import preprocess_input from skimage.metrics import structural_similarity as ssim from PIL import Image from io import BytesIO # Disable GPU for TensorFlow os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0" os.environ["CUDA_VISIBLE_DEVICES"] = "-1" # --- DOCUMENTATION STRINGS (English Only) --- GUIDELINE_SETUP = """ ## 1. Quick Start Guide: Setup and Run Instructions This application uses a combination of advanced feature extraction (ResNet50) and structural analysis (SSIM) to determine if comparison images are structurally and semantically similar to a reference image. 1. **Upload Reference:** Upload the main image you want to compare against in the 'Reference Image' box. 2. **Upload Comparisons:** Upload one or more images you want to test for similarity in the 'Comparison Images' file upload area. 3. **Set Threshold:** Adjust the 'Similarity Threshold' slider. This controls the sensitivity for structural similarity (SSIM). 4. **Run:** Click the **"Compare Images"** button. 5. **Review:** Results will appear in the 'Results' panel, indicating if each comparison image is "SIMILAR" or "NOT SIMILAR". """ GUIDELINE_INPUT = """ ## 2. Expected Inputs and Preprocessing | Input Field | Purpose | Requirement | | :--- | :--- | :--- | | **Reference Image** | The baseline image against which all others will be compared. | Must be a single image file (JPG, PNG). | | **Comparison Images** | One or more images to be tested for similarity. | Must be multiple image files. Upload them using the file selector. | | **Similarity Threshold** | A slider controlling the sensitivity (0.0 to 1.0) for structural similarity (SSIM). | Higher values (closer to 1.0) mean stricter similarity requirements. Default is 0.5. | **Image Preprocessing:** All uploaded images are automatically resized to 224x224 pixels and standardized according to the requirements of the ResNet model before feature extraction. """ GUIDELINE_OUTPUT = """ ## 3. Expected Outputs (Similarity Results) The application provides two main outputs: 1. **Results (HTML Panel):** * A list detailing the outcome for each comparison image. * Status: **SIMILAR** (Green) or **NOT SIMILAR** (Red). 2. **Processed Images (Gallery):** * A gallery displaying the input comparison images after they have been processed. ### How Similarity is Determined: The classification relies on two checks: Structural Similarity (SSIM) and Deep Feature Distance (ResNet). An image is marked "SIMILAR" if both structural and semantic properties suggest a close match. """ # --- CLASSIFIER CLASS --- class ImageCharacterClassifier: def __init__(self, similarity_threshold=0.5): self.model = ResNet50(weights='imagenet', include_top=False, pooling='avg') self.similarity_threshold = similarity_threshold def load_and_preprocess_image(self, img): img = img.convert('RGB') img_array = np.array(img) img_array = cv2.resize(img_array, (224, 224)) img_array = np.expand_dims(img_array, axis=0) img_array = preprocess_input(img_array) return img_array def extract_features(self, img): preprocessed_img = self.load_and_preprocess_image(img) features = self.model.predict(preprocessed_img, verbose=0) return features def calculate_ssim(self, img1, img2): img1_gray = cv2.cvtColor(img1, cv2.COLOR_RGB2GRAY) img2_gray = cv2.cvtColor(img2, cv2.COLOR_RGB2GRAY) img2_gray = cv2.resize(img2_gray, (img1_gray.shape[1], img1_gray.shape[0])) return ssim(img1_gray, img2_gray, data_range=img1_gray.max() - img1_gray.min()) def process_images(reference_image_array, comparison_files, similarity_threshold): try: if reference_image_array is None: return "

Please upload a reference image.

", [] if not comparison_files: return "

Please upload comparison images.

", [] classifier = ImageCharacterClassifier(similarity_threshold) ref_image_pil = Image.fromarray(reference_image_array).convert("RGB") ref_features = classifier.extract_features(ref_image_pil) ref_image_for_ssim = cv2.cvtColor(reference_image_array, cv2.COLOR_BGR2RGB) results = [] html_output = "

Comparison Results:

" for comp_file_item in comparison_files: try: # FIX: Extract file path correctly regardless of whether it's a dict (internal Gradio) # or a gr.File object (returned by our custom loader function). if isinstance(comp_file_item, str): file_path = comp_file_item elif hasattr(comp_file_item, 'name'): file_path = comp_file_item.name elif isinstance(comp_file_item, dict) and 'name' in comp_file_item: file_path = comp_file_item['name'] else: raise ValueError("Invalid file object structure.") with open(file_path, "rb") as f: comp_pil = Image.open(BytesIO(f.read())).convert("RGB") comp_array = np.array(comp_pil) # SSIM Check ssim_score = classifier.calculate_ssim(ref_image_for_ssim, comp_array) # Feature Check comp_features = classifier.extract_features(comp_pil) max_feature_diff = np.max(np.abs(ref_features - comp_features)) feature_match = max_feature_diff < 6.0 is_similar = feature_match # Primary criterion status_text = f"SIMILAR (SSIM: {ssim_score:.3f})" if is_similar else f"NOT SIMILAR (SSIM: {ssim_score:.3f})" status_color = "green" if is_similar else "red" html_output += f"

{os.path.basename(file_path)}: {status_text}

" results.append(comp_array) except Exception as e: # Use the path for logging the error error_name = os.path.basename(file_path) if 'file_path' in locals() else 'Unknown File' html_output += f"

Error processing {error_name}: {str(e)}

" return html_output, [r for r in results if r is not None] except Exception as e: return f"

Critical Error: {str(e)}

", [] # --- SAMPLE DATA DEFINITION --- # Placeholder file paths (MUST EXIST for examples to work) # NOTE: Adjusted paths to match your provided snippet structure 'sample_data/filename' SAMPLE_FILES_SET1 = { "reference": "sample_data/license3.jpg", "comparisons": ["sample_data/license3.jpg", "sample_data/license3.jpg", "sample_data/licence.jpeg"] } SAMPLE_FILES_SET2 = { "reference": "sample_data/licence.jpeg", "comparisons": ["sample_data/licence.jpeg", "sample_data/license3.jpg", "sample_data/licence.jpeg", "sample_data/licence.jpeg"] } # --- GRADIO UI SETUP --- def create_interface(): with gr.Blocks(title="Image Similarity Classifier") as interface: gr.Markdown("# Image Similarity Classifier (ResNet + SSIM)") gr.Markdown("Tool to compare a reference image against multiple comparison images based on structural and deep feature similarity.") # 1. Guidelines Section with gr.Accordion("User Guidelines and Documentation", open=False): gr.Markdown(GUIDELINE_SETUP) gr.Markdown("---") gr.Markdown(GUIDELINE_INPUT) gr.Markdown("---") gr.Markdown(GUIDELINE_OUTPUT) gr.Markdown("---") # 2. Application Interface with gr.Row(): with gr.Column(): gr.Markdown("## Step 1: Upload a Reference Image ") reference_input = gr.Image(label="Reference Image", type="numpy", height=300) gr.Markdown("## Step 2: Upload Multiple Images to Compair with Reference Image ") comparison_input = gr.Files(label="Comparison Images", type="file") gr.Markdown("## Step 3: Set the Confidence Score (Optional) ") threshold_slider = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, step=0.05, label="Similarity Threshold (SSIM)") gr.Markdown("## Step 4: Click Compare Images ") submit_button = gr.Button("Compare Images", variant="primary") gr.Markdown("---") gr.Markdown("# Results ") gr.Markdown("## Comparison Result ") output_html = gr.HTML(label="Comparison Results") gr.Markdown("## Processed Comparison Images") output_gallery = gr.Gallery(label="Processed Comparison Images", columns=3) # 3. Example Loading Setup gr.Markdown("---") gr.Markdown("## Sample Data for Testing") gr.Markdown("### Click on any of these two set to run the test set ") def load_and_run_set(reference_path, comparison_paths, threshold_value=0.5): """Loads data into inputs, triggers processing, and returns all results.""" # 1. Load Reference Image as NumPy array ref_img_pil = Image.open(reference_path).convert("RGB") ref_img_array = np.array(ref_img_pil) # 2. Comparison Files: Prepare the list of paths (strings) for the processor # We return a list of strings/paths here, which Gradio's gr.Files component accepts comparison_file_paths = comparison_paths # 3. Process the images immediately using the paths html, gallery = process_images(ref_img_array, comparison_file_paths, threshold_value) # 4. Return inputs and outputs for component update return ref_img_array, comparison_file_paths, threshold_value, html, gallery with gr.Row(): btn_set1 = gr.Button("Load & Run Sample Set 1 (Similar Docs)", size="sm") btn_set2 = gr.Button("Load & Run Sample Set 2 (Dissimilar Docs)", size="sm") # 4. Event Handling submit_button.click( fn=process_images, inputs=[reference_input, comparison_input, threshold_slider], outputs=[output_html, output_gallery] ) # Event handlers for example buttons: load data into inputs/outputs btn_set1.click( fn=lambda: load_and_run_set(SAMPLE_FILES_SET1['reference'], SAMPLE_FILES_SET1['comparisons'], 0.6), inputs=[], outputs=[reference_input, comparison_input, threshold_slider, output_html, output_gallery] ) btn_set2.click( fn=lambda: load_and_run_set(SAMPLE_FILES_SET2['reference'], SAMPLE_FILES_SET2['comparisons'], 0.4), inputs=[], outputs=[reference_input, comparison_input, threshold_slider, output_html, output_gallery] ) return interface if __name__ == "__main__": # Ensure the 'sample_data/' directory exists with 'license3.jpg' and 'licence.jpeg' # and any other necessary files. interface = create_interface() interface.queue() interface.launch( server_name="0.0.0.0", server_port=7860 )