phitran/image-processing

1import gradio as gr2import cv23import numpy as np4from PIL import Image5from ultralytics import YOLO  # YOLOv8 from Ultralytics6 7# Load YOLOv8 model (pre-trained on COCO dataset)8model = YOLO("yolov8n.pt")  # Using the "nano" model (fast & lightweight)9 10#apply smoothing using OpenCV's medianBlur11def smooth_image(image):12    image = np.array(image)  # Convert PIL image to NumPy array13    smoothed = cv2.medianBlur(image, 15)  # Apply median blur with kernel size 514    return Image.fromarray(smoothed)  # Convert back to PIL image15 16#apply Erosion Morphological Transformation17def erode_image(image):18    image = np.array(image)19    kernel = np.ones((3, 3), np.uint8)  # Define a 3x3 kernel20    eroded = cv2.erode(image, kernel, iterations=1)  # Apply erosion21    return Image.fromarray(eroded)  # Convert back to PIL image22 23#apply image segmentation using Otsu's Thresholding24def segment_image(image):25    image = np.array(image)26    gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)  # Convert to grayscale27    _, segmented = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)  # Apply Otsu's thresholding28    return Image.fromarray(segmented)  # Convert back to PIL image29 30#apply Fourier Transform and display magnitude spectrum31def fourier_transform(image):32    image = np.array(image)33    gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)  # Convert to grayscale34    dft = np.fft.fft2(gray)  # Compute Fourier Transform35    dft_shift = np.fft.fftshift(dft)  # Shift zero frequency to center36    magnitude_spectrum = 20 * np.log(np.abs(dft_shift) + 1)  # Compute magnitude spectrum37    magnitude_spectrum = np.uint8(255 * (magnitude_spectrum / np.max(magnitude_spectrum)))  # Normalize for display38    return Image.fromarray(magnitude_spectrum)39 40def detect_objects(image):41    image = np.array(image)  # Convert PIL image to NumPy array42 43    # Perform object detection44    results = model(image)45 46    # Process detections47    for result in results:48        boxes = result.boxes.xyxy  # Bounding boxes (x1, y1, x2, y2)49        confidences = result.boxes.conf  # Confidence scores50        class_ids = result.boxes.cls.int().tolist()  # Class labels51 52        for box, conf, class_id in zip(boxes, confidences, class_ids):53            x1, y1, x2, y2 = map(int, box.tolist())54            label = f"{model.names[class_id]} ({conf:.2f})"55 56            # Draw bounding box & label57            cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0), 2)58            cv2.putText(image, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)59 60    return Image.fromarray(image)  # Convert back to PIL Image for Gradio61 62 63 64def create_interface():65    with gr.Blocks() as demo:66        with gr.Row():67            with gr.Column():68                image_input = gr.Image(label="Upload Image", type="pil")69            with gr.Column():70                output_image = gr.Image(label="Processed Image", type="pil")71 72        with gr.Row():73            smoothing_button = gr.Button("Smoothing/ Blurring")74            morphological_transform_button = gr.Button("Morphological Transformations")75            fourier_transform_button = gr.Button("Fourier Transform")76            segmentation_button = gr.Button("Segmentation")77            object_recognition_button = gr.Button("Object Recognition (YOLO)")78 79        # Link buttons to their respective functions80        smoothing_button.click(smooth_image, inputs=image_input, outputs=output_image)81        morphological_transform_button.click(erode_image, inputs=image_input, outputs=output_image)82        fourier_transform_button.click(fourier_transform, inputs=image_input, outputs=output_image)83        segmentation_button.click(segment_image, inputs=image_input, outputs=output_image)84        object_recognition_button.click(detect_objects, inputs=image_input, outputs=output_image)85 86    return demo87 88 89# Launch the Gradio app90app = create_interface()91app.launch()92