import gradio as gr import numpy as np from PIL import Image from skimage.util import random_noise import cv2 # Assuming ImageSlider is a custom or extended component of Gradio from gradio_imageslider import ImageSlider # Function to add noise to the image def add_noise(image, noise_type, mean=0, var=0.01, amount=0.05, salt_vs_pepper=0.5): # Convert image to float for processing image = np.array(image).astype(float) / 255.0 # Normalize the image kwargs = {} # Set noise parameters based on the selected noise type if noise_type in ['gaussian', 'speckle']: kwargs['mean'] = mean kwargs['var'] = var elif noise_type in ['salt', 'pepper', 's&p']: kwargs['amount'] = amount if noise_type == 's&p': kwargs['salt_vs_pepper'] = salt_vs_pepper elif noise_type == 'localvar': kwargs['local_vars'] = np.full(image.shape, var) # Add noise to the image noisy_image = random_noise(image, mode=noise_type.replace("s&p", "salt&pepper"), **kwargs, clip=True) return Image.fromarray((noisy_image * 255).astype(np.uint8)) # Function to apply denoising to the image def apply_denoising(image, method, gaussian_kernel, median_kernel, bilateral_diameter, bilateral_sigma_color, bilateral_sigma_space, nlm_h, nlm_template_window_size, nlm_search_window_size): # Convert image to array for processing image = np.array(image) # Apply the selected denoising method if method == "Gaussian Blur": denoised = cv2.GaussianBlur(image, (gaussian_kernel, gaussian_kernel), 0) elif method == "Median Blur": denoised = cv2.medianBlur(image, median_kernel) elif method == "Bilateral Filter": denoised = cv2.bilateralFilter(image, bilateral_diameter, bilateral_sigma_color, bilateral_sigma_space) elif method == "Non-Local Means": denoised = cv2.fastNlMeansDenoisingColored(image, None, nlm_h, nlm_h, nlm_template_window_size, nlm_search_window_size) return Image.fromarray(denoised) # Function to apply morphological operations def apply_morphological_operation(image, kernel_size, iterations, operation): image = np.array(image) kernel = np.ones((kernel_size, kernel_size), np.uint8) if operation == "Erosion": result = cv2.erode(image, kernel, iterations=iterations) elif operation == "Dilation": result = cv2.dilate(image, kernel, iterations=iterations) elif operation == "Opening": result = cv2.morphologyEx(image, cv2.MORPH_OPEN, kernel, iterations=iterations) elif operation == "Closing": result = cv2.morphologyEx(image, cv2.MORPH_CLOSE, kernel, iterations=iterations) return Image.fromarray(result) # Function to apply edge detection def apply_edge_detection(image, min_val, max_val, operation, kernel_size): image = np.array(image.convert('L')) if operation == "Canny": edges = cv2.Canny(image, min_val, max_val) elif operation == "Sobel-X": edges = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=kernel_size) elif operation == "Sobel-Y": edges = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=kernel_size) elif operation == "Sobel-XY": edges_x = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=kernel_size) edges_y = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=kernel_size) edges = cv2.addWeighted(edges_x, 0.5, edges_y, 0.5, 0) elif operation == "Laplacian": edges = cv2.Laplacian(image, cv2.CV_64F, ksize=kernel_size) edges = np.clip(edges, 0, 255).astype(np.uint8) return Image.fromarray(edges) # Gradio interface setup with gr.Blocks() as demo: gr.Markdown("# OpenCV Image Processing with Gradio - Add Noise, Remove Noise, Morphological Operations and Edge Detection") tab_names = ["Add Noise", "Remove Noise", "Morphological Operations", "Edge Detection"] # ---- ADD NOISE TAB ---- with gr.Tab("Add Noise"): with gr.Row(): img_input = gr.Image(label="Input Image", type="pil") img_output = gr.Image(label="Output Image", type="pil") noise_type = gr.Radio(["gaussian", "localvar", "poisson", "salt", "pepper", "s&p", "speckle"], label="Type of Noise", value="gaussian") mean_slider = gr.Slider(0, 1, value=0, label="Mean (for Gaussian/Speckle)", visible=True) var_slider = gr.Slider(0, 0.1, value=0.01, label="Variance", visible=True) amount_slider = gr.Slider(0, 1, value=0.05, label="Amount (for Salt/Pepper/S&P)", visible=False) salt_vs_pepper_slider = gr.Slider(0, 1, value=0.5, label="Salt vs Pepper (for S&P)", visible=False) noise_button = gr.Button("Add Noise") def on_noise_type_change(noise_type): if noise_type in ['gaussian', 'speckle']: return gr.update(visible=True), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False) elif noise_type in ['salt', 'pepper']: return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True), gr.update(visible=False) elif noise_type == 's&p': return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True), gr.update(visible=True) elif noise_type == 'localvar': return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False) noise_type.change(fn=on_noise_type_change, inputs=noise_type, outputs=[mean_slider, var_slider, amount_slider, salt_vs_pepper_slider]) noise_button.click(fn=add_noise, inputs=[img_input, noise_type, mean_slider, var_slider, amount_slider, salt_vs_pepper_slider], outputs=img_output) # ---- REMOVE NOISE TAB ---- with gr.Tab("Remove Noise"): with gr.Row(): denoise_img_input = gr.Image(label="Input Noisy Image", type="pil") denoise_img_output = gr.Image(label="Output Image", type="pil") denoise_method = gr.Radio(["Gaussian Blur", "Median Blur", "Bilateral Filter", "Non-Local Means"], label="Denoising Method", value="Gaussian Blur") gaussian_kernel = gr.Slider(1, 31, step=2, value=5, label="Gaussian Kernel Size", visible=True) median_kernel = gr.Slider(1, 31, step=2, value=5, label="Median Kernel Size", visible=False) bilateral_diameter = gr.Slider(1, 31, step=2, value=9, label="Bilateral Filter Diameter", visible=False) bilateral_sigma_color = gr.Slider(1, 150, value=75, label="Bilateral Filter Sigma Color", visible=False) bilateral_sigma_space = gr.Slider(1, 150, value=75, label="Bilateral Filter Sigma Space", visible=False) nlm_h = gr.Slider(1, 20, value=10, label="Non-Local Means h", visible=False) nlm_template_window_size = gr.Slider(1, 21, step=2, value=7, label="Non-Local Means Template Window Size", visible=False) nlm_search_window_size = gr.Slider(1, 51, step=2, value=21, label="Non-Local Means Search Window Size", visible=False) denoise_button = gr.Button("Remove Noise") def on_denoise_method_change(method): if method == "Gaussian Blur": return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False) elif method == "Median Blur": return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False) elif method == "Bilateral Filter": return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True), gr.update(visible=True), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False) elif method == "Non-Local Means": return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=True), gr.update(visible=True), gr.update(visible=True) denoise_method.change(fn=on_denoise_method_change, inputs=denoise_method, outputs=[gaussian_kernel, median_kernel, bilateral_diameter, bilateral_sigma_color, bilateral_sigma_space, nlm_h, nlm_template_window_size, nlm_search_window_size]) denoise_button.click(fn=apply_denoising, inputs=[denoise_img_input, denoise_method, gaussian_kernel, median_kernel, bilateral_diameter, bilateral_sigma_color, bilateral_sigma_space, nlm_h, nlm_template_window_size, nlm_search_window_size], outputs=denoise_img_output) # ---- MORPHOLOGICAL OPERATIONS TAB ---- with gr.Tab("Morphological Operations"): with gr.Row(): morph_img_input = gr.Image(label="Input Image", type="pil") morph_img_output = gr.Image(label="Output Image", type="pil") kernel_slider = gr.Slider(1, 11, value=3, step=2, label="Kernel Size") iter_slider = gr.Slider(1, 10, value=1, step=1, label="Iterations") morph_operation = gr.Radio(["Erosion", "Dilation", "Opening", "Closing"], label="Morphological Operation", value="Erosion") apply_morph_button = gr.Button("Apply Morphological Operation") apply_morph_button.click(fn=apply_morphological_operation, inputs=[morph_img_input, kernel_slider, iter_slider, morph_operation], outputs=morph_img_output) # ---- EDGE DETECTION TAB ---- with gr.Tab("Edge Detection"): with gr.Row(): edge_img_input = gr.Image(label="Input Image", type="pil") edge_img_output = gr.Image(label="Output Image", type="pil") min_val_slider = gr.Slider(50, 150, label="Min Threshold", visible=True) max_val_slider = gr.Slider(100, 200, label="Max Threshold", visible=True) kernel_size_slider = gr.Slider(1, 11, value=3, step=2, label="Kernel Size", visible=True) edge_operation = gr.Radio(["Canny", "Sobel-X", "Sobel-Y", "Sobel-XY", "Laplacian"], label="Edge Operation", value="Canny") apply_edge_button = gr.Button("Apply Edge Detection") def on_edge_operation_change(operation): if operation == "Canny": return gr.update(visible=True), gr.update(visible=True), gr.update(visible=False) else: return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True) edge_operation.change(fn=on_edge_operation_change, inputs=edge_operation, outputs=[min_val_slider, max_val_slider, kernel_size_slider]) apply_edge_button.click(fn=apply_edge_detection, inputs=[edge_img_input, min_val_slider, max_val_slider, edge_operation, kernel_size_slider], outputs=edge_img_output) # ---- DYNAMIC TRANSFER BUTTON ---- with gr.Row(): source_tab_dropdown = gr.Dropdown(tab_names, label="Transfer From Tab") destination_tab_dropdown = gr.Dropdown(tab_names, label="Transfer To Tab") transfer_image_button = gr.Button("Transfer Image") def dynamic_image_transfer(add_noise_input, add_noise_output, denoise_input, denoise_output, morph_input, morph_output, edge_input, edge_output, source, destination): image_to_send = None if source == "Add Noise": image_to_send = add_noise_output if add_noise_output else add_noise_input elif source == "Remove Noise": image_to_send = denoise_output if denoise_output else denoise_input elif source == "Morphological Operations": image_to_send = morph_output if morph_output else morph_input elif source == "Edge Detection": image_to_send = edge_output if edge_output else edge_input updates = { "Add Noise": gr.update(value=image_to_send) if destination == "Add Noise" else gr.update(), "Remove Noise": gr.update(value=image_to_send) if destination == "Remove Noise" else gr.update(), "Morphological Operations": gr.update(value=image_to_send) if destination == "Morphological Operations" else gr.update(), "Edge Detection": gr.update(value=image_to_send) if destination == "Edge Detection" else gr.update(), } return [updates.get("Add Noise"), updates.get("Remove Noise"), updates.get("Morphological Operations"), updates.get("Edge Detection")] transfer_image_button.click( fn=dynamic_image_transfer, inputs=[img_input, img_output, denoise_img_input, denoise_img_output, morph_img_input, morph_img_output, edge_img_input, edge_img_output, source_tab_dropdown, destination_tab_dropdown], outputs=[img_input, denoise_img_input, morph_img_input, edge_img_input] ) # Launch the Gradio interface demo.launch()