added App

app.py

@@ -0,0 +1,412 @@
+import streamlit as st
+import cv2
+import numpy as np
+from PIL import Image
+import time
+from streamlit_drawable_canvas import st_canvas
+import matplotlib.pylab as plt
+from estimate_homography import calculate_homography, fit_image_in_target_space
+
+stitched_image_rgb, stitched_result = None, None
+
+# Function to load an image from uploaded file
+def load_image(uploaded_file):
+    img = cv2.imdecode(np.frombuffer(uploaded_file.read(), np.uint8), cv2.IMREAD_GRAYSCALE)
+    return img
+
+# Function to compute stereo vision and disparity map
+def compute_stereo_vision(img1, img2):
+    # Feature Detection and Matching using ORB (ORB is a good alternative for uncalibrated cameras)
+    orb = cv2.ORB_create()  # ORB is a good alternative to SIFT for uncalibrated cameras
+    kp1, des1 = orb.detectAndCompute(img1, None)
+    kp2, des2 = orb.detectAndCompute(img2, None)
+
+    # BFMatcher with default params
+    bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
+    matches = bf.match(des1, des2)
+
+    # Sort matches by distance
+    matches = sorted(matches, key=lambda x: x.distance)
+
+    # Estimate the Fundamental Matrix
+    pts1 = np.array([kp1[m.queryIdx].pt for m in matches])
+    pts2 = np.array([kp2[m.trainIdx].pt for m in matches])
+
+    # Fundamental matrix using RANSAC to reject outliers
+    F, mask = cv2.findFundamentalMat(pts1, pts2, cv2.FM_RANSAC)
+
+    # Estimate the Camera Pose (Rotation and Translation)
+    K = np.eye(3)  # Assuming no camera calibration
+    E = K.T @ F @ K  # Essential matrix
+    _, R, T, _ = cv2.recoverPose(E, pts1, pts2)
+
+    # Stereo Rectification
+    stereo_rectify = cv2.stereoRectify(K, None, K, None, img1.shape[::-1], R, T, alpha=0)
+    left_map_x, left_map_y = cv2.initUndistortRectifyMap(K, None, R, K, img1.shape[::-1], cv2.CV_32F)
+    right_map_x, right_map_y = cv2.initUndistortRectifyMap(K, None, R, K, img2.shape[::-1], cv2.CV_32F)
+
+    # Apply the rectification transformations to the images
+    img1_rectified = cv2.remap(img1, left_map_x, left_map_y, interpolation=cv2.INTER_LINEAR)
+    img2_rectified = cv2.remap(img2, right_map_x, right_map_y, interpolation=cv2.INTER_LINEAR)
+
+    # Resize img2_rectified to match img1_rectified size (if necessary)
+    if img1_rectified.shape != img2_rectified.shape:
+        img2_rectified = cv2.resize(img2_rectified, (img1_rectified.shape[1], img1_rectified.shape[0]))
+
+    # Disparity Map Computation using StereoBM
+    stereo = cv2.StereoBM_create(numDisparities=16, blockSize=15)
+    disparity = stereo.compute(img1_rectified, img2_rectified)
+
+    return disparity, img1_rectified, img2_rectified
+
+
+def run_point_est(world_pts, img_pts, img):
+    if isinstance(img_pts, list):
+        img_pts = np.array(img_pts)
+
+    if isinstance(world_pts, list):
+        world_pts = np.array(world_pts)
+
+    # Plot the original image with marked points
+    st.write("Original Image with Points")
+    plt.figure()
+    plt.imshow(img)
+    plt.scatter(img_pts[:, 0], img_pts[:, 1], color='red')
+    plt.axis("off")
+    plt.title("Original Image with img points marked in red")
+    st.pyplot(plt)
+
+    H = calculate_homography(img_pts, world_pts)  # img_pts = H * world_pts
+
+    #### Cross check ####
+    t_one = np.ones((img_pts.shape[0], 1))
+    t_out_pts = np.concatenate((world_pts, t_one), axis=1)
+    x = np.matmul(H, t_out_pts.T)
+    x = x / x[-1, :]
+
+    st.write("Given Image Points:", img_pts)
+    st.write("Calculated Image Points:", x.T)
+    st.write("Homography Matrix (OpenCV):", cv2.findHomography(world_pts, img_pts)[0])
+    st.write("Calculated Homography Matrix:", H)
+
+    #####################
+    h, w, _ = img.shape
+    corners_img = np.array([[0, 0], [w, 0], [w, h], [0, h]])
+    H_inv = np.linalg.inv(H)
+    t_out_pts = np.concatenate((corners_img, t_one), axis=1)
+    world_crd_corners = np.matmul(H_inv, t_out_pts.T)
+    world_crd_corners = world_crd_corners / world_crd_corners[-1, :]  # Normalize
+
+    min_crd = np.amin(world_crd_corners.T, axis=0)
+    max_crd = np.amax(world_crd_corners.T, axis=0)
+
+    offset = min_crd.astype(np.int64)
+    offset[2] = 0
+
+    width_world = np.ceil(max_crd - min_crd)[0] + 1
+    height_world = np.ceil(max_crd - min_crd)[1] + 1
+
+    world_img = np.zeros((int(height_world), int(width_world), 3), dtype=np.uint8)
+    mask = np.ones((int(height_world), int(width_world)))
+
+    out = fit_image_in_target_space(img, world_img, mask, H, offset)
+
+    st.write("Corrected Image")
+    plt.figure()
+    plt.imshow(out)
+    plt.axis("off")
+    plt.title("Corrected Image with Point Point Correspondence")
+    st.pyplot(plt)
+
+
+# Function to stitch images
+def stitch_images(images):
+    stitcher = cv2.Stitcher_create() if cv2.__version__.startswith('4') else cv2.createStitcher()
+    status, stitched_image = stitcher.stitch(images)
+    if status == cv2.Stitcher_OK:
+        return stitched_image, status
+    else:
+        return None, status
+
+# Function to match features
+def match_features(images):
+    if len(images) < 2:
+        return None, "At least two images are required for feature matching."
+    
+    gray1 = cv2.cvtColor(images[0], cv2.COLOR_BGR2GRAY)
+    gray2 = cv2.cvtColor(images[1], cv2.COLOR_BGR2GRAY)
+    
+    sift = cv2.SIFT_create()
+    keypoints1, descriptors1 = sift.detectAndCompute(gray1, None)
+    keypoints2, descriptors2 = sift.detectAndCompute(gray2, None)
+    
+    bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
+    matches = bf.match(descriptors1, descriptors2)
+    matches = sorted(matches, key=lambda x: x.distance)
+    
+    matched_image = cv2.drawMatches(images[0], keypoints1, images[1], keypoints2, matches[:50], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
+    return matched_image, None
+
+# Function to cartoonify an image
+def cartoonify_image(image):
+    # Convert to grayscale
+    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+
+    gray_blur = cv2.medianBlur(gray, 7)
+
+    edges = cv2.adaptiveThreshold(
+        gray_blur, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 9, 10
+    )
+
+    color = cv2.bilateralFilter(image, 9, 250, 250)
+
+    cartoon = cv2.bitwise_and(color, color, mask=edges)
+
+    return cartoon
+
+# Streamlit layout and UI
+st.set_page_config(page_title="Image Stitching and Feature Matching", layout="wide")
+st.title("Image Stitching and Feature Matching Application")
+
+# State to store captured images
+if "captured_images" not in st.session_state:
+    st.session_state["captured_images"] = []
+
+if "stitched_image" not in st.session_state:
+    st.session_state["stitched_image"] = None
+# Sidebar for displaying captured images
+st.sidebar.header("Captured Images")
+if st.session_state["captured_images"]:
+    placeholder = st.sidebar.empty()
+    with placeholder.container():
+        for i, img in enumerate(st.session_state["captured_images"]):
+            img_thumbnail = cv2.resize(img, (100, 100))
+            st.image(cv2.cvtColor(img_thumbnail, cv2.COLOR_BGR2RGB), caption=f"Image {i+1}", use_container_width =False)
+            if st.button(f"Delete Image {i+1}", key=f"delete_{i}"):
+                st.session_state["captured_images"].pop(i)
+                placeholder.empty()  # Clear and refresh the sidebar
+                break
+
+# Capture the image from camera input
+st.header("Upload or Capture Images")
+uploaded_files = st.file_uploader("Upload images", type=["jpg", "jpeg", "png"], accept_multiple_files=True)
+captured_image = st.camera_input("Take a picture using your camera")
+
+if st.button("Add Captured Image"):
+    if captured_image:
+        captured_image_array = cv2.cvtColor(np.array(Image.open(captured_image)), cv2.COLOR_RGB2BGR)
+        st.session_state["captured_images"].append(captured_image_array)
+        st.success(f"Captured image {len(st.session_state['captured_images'])} added!")
+
+# Combine uploaded and captured images
+images = [cv2.cvtColor(np.array(Image.open(file)), cv2.COLOR_RGB2BGR) for file in uploaded_files]
+images.extend(st.session_state["captured_images"])
+
+st.write(f"Total images: {len(images)}")
+
+# Placeholder for dynamic updates
+loading_placeholder = st.empty()
+
+# Function to show the loading animation
+def show_loading_bar(placeholder):
+    with placeholder:
+        st.write("Processing images... Please wait.")
+        time.sleep(2)
+
+if st.button("Stitch Images"):
+    if len(images) < 2:
+        st.error("Please provide at least two images for stitching.")
+    else:
+        show_loading_bar(loading_placeholder)
+        stitched_result, status = stitch_images(images)
+        loading_placeholder.empty()
+        if stitched_result is not None:
+            stitched_image_rgb = cv2.cvtColor(stitched_result, cv2.COLOR_BGR2RGB)
+            st.image(stitched_image_rgb, caption="Stitched Image", use_container_width=True)
+            st.session_state["stitched_image"] = stitched_image_rgb
+            st.success("Stitching completed successfully!")
+        else:
+            st.error(f"Stitching failed with status: {status}.")
+if st.button("Show Matching Features"):
+    if len(images) < 2:
+        st.error("Please provide at least two images for feature matching.")
+    else:
+        show_loading_bar(loading_placeholder)
+        matched_image, error = match_features(images)
+        loading_placeholder.empty()
+        if matched_image is not None:
+            matched_image_rgb = cv2.cvtColor(matched_image, cv2.COLOR_BGR2RGB)
+            st.image(matched_image_rgb, caption="Feature Matching Visualization", use_container_width=True)
+            st.success("Feature matching completed successfully!")
+        else:
+            st.error(error)
+
+if st.session_state["stitched_image"] is not None:
+    st.header("Homography Transformation on Stitched Image")
+
+    st.write("### Select Points on Stitched Image")
+    stitched_image = st.session_state["stitched_image"]
+    image = Image.fromarray(cv2.cvtColor(stitched_image, cv2.COLOR_BGR2RGB))
+
+    canvas_result = st_canvas(
+        fill_color="rgba(255, 0, 0, 0.3)",
+        stroke_width=3,
+        background_image=image,
+        update_streamlit=True,
+        drawing_mode="point",
+        height=image.height,
+        width=image.width,
+        key="canvas",
+    )
+
+    img_pts = []
+
+    if canvas_result.json_data is not None:
+        for obj in canvas_result.json_data["objects"]:
+            if obj["type"] == "circle":
+                x = obj["left"] + obj["width"] / 2
+                y = obj["top"] + obj["height"] / 2
+                img_pts.append([int(x), int(y)])
+
+    if img_pts:
+        st.write("### Selected Image Points")
+        st.write(img_pts)
+
+        st.write("### Enter Corresponding World Points")
+        world_pts = st.text_area(
+            "Enter world points as a list of tuples (e.g., [(0, 0), (300, 0), (0, 400), (300, 400)])",
+            value="[(0, 0), (300, 0), (0, 400), (300, 400)]",
+        )
+
+        if st.button("Run Homography Transformation"):
+            try:
+                world_pts = eval(world_pts)
+                if len(world_pts) != len(img_pts):
+                    st.error("The number of world points must match the number of image points.")
+                else:
+                    run_point_est(world_pts, img_pts, stitched_image)
+            except Exception as e:
+                st.error(f"Error: {e}")
+
+
+if "stitched_image" in st.session_state:
+    st.header("Cartoonify & Do Homography on Your Stitched Image")
+    if st.button("Cartoonify Stitched Image"):
+        cartoon = cartoonify_image(cv2.cvtColor(st.session_state["stitched_image"], cv2.COLOR_RGB2BGR))
+        st.image(cv2.cvtColor(cartoon, cv2.COLOR_BGR2RGB), caption="Cartoonified Image", use_container_width=True)
+        st.success("Cartoonification completed successfully!")
+
+# Upload images
+st.subheader("Upload Left and Right Images")
+left_image_file = st.file_uploader("Choose the Left Image", type=["jpg", "png", "jpeg"])
+right_image_file = st.file_uploader("Choose the Right Image", type=["jpg", "png", "jpeg"])
+
+# Check if both images are uploaded
+if left_image_file and right_image_file:
+    # Load the uploaded images
+    img1 = load_image(left_image_file)
+    img2 = load_image(right_image_file)
+
+    # Display the uploaded images
+    st.image(img1, caption="Left Image", use_container_width =True)
+    st.image(img2, caption="Right Image", use_container_width =True)
+
+    # Compute the stereo vision and disparity map
+    disparity, img1_rectified, img2_rectified = compute_stereo_vision(img1, img2)
+
+    # Display the rectified images
+    # st.subheader("Rectified Left Image")
+    # st.image(img1_rectified, caption="Rectified Left Image", use_container_width =True)
+
+    # st.subheader("Rectified Right Image")
+    # st.image(img2_rectified, caption="Rectified Right Image", use_container_width =True)
+
+    # Show the disparity map
+    fig, ax = plt.subplots()
+    st.subheader("Disparity Map")
+    plt.imshow(disparity, cmap='gray')
+    plt.title("Disparity Map")
+    plt.colorbar()
+    st.pyplot(fig)
+
+    # # Optionally: Display an anaglyph or combined view of the images
+    # anaglyph = cv2.merge([img1_rectified, np.zeros_like(img1_rectified), img2_rectified])
+    # st.subheader("Anaglyph Stereo View")
+    # st.image(anaglyph, caption="Anaglyph Stereo View", use_container_width =True)
+
+
+
+# if "img_pts" not in st.session_state:
+#     st.session_state["img_pts"] = []
+
+# if "world_pts" not in st.session_state:
+#     st.session_state["world_pts"] = []
+
+# if "homography_ready" not in st.session_state:
+#     st.session_state["homography_ready"] = False
+
+# if st.button('Homography Transformation'):
+#     if st.session_state["stitched_image"] is not None:
+#         st.write("### Select Points on Stitched Image")
+#         stitched_image = st.session_state["stitched_image"]
+#         image = Image.fromarray(cv2.cvtColor(stitched_image, cv2.COLOR_BGR2RGB))
+
+#         # Display canvas for selecting points
+#         canvas_result = st_canvas(
+#             fill_color="rgba(255, 0, 0, 0.3)",
+#             stroke_width=3,
+#             background_image=image,
+#             update_streamlit=True,
+#             drawing_mode="point",
+#             height=image.height,
+#             width=image.width,
+#             key="canvas",
+#         )
+
+#         # Collect selected points
+#         if canvas_result.json_data is not None:
+#             img_pts_temp = []
+#             for obj in canvas_result.json_data["objects"]:
+#                 if obj["type"] == "circle":
+#                     x = obj["left"] + obj["width"] / 2
+#                     y = obj["top"] + obj["height"] / 2
+#                     img_pts_temp.append([int(x), int(y)])
+
+#             # Only update points if there are new ones
+#             if img_pts_temp:
+#                 st.session_state["img_pts"] = img_pts_temp
+
+#         # Display the selected points
+#         if st.session_state["img_pts"]:
+#             st.write("### Selected Image Points")
+#             st.write(st.session_state["img_pts"])
+
+#             # Input world points
+#             world_pts_input = st.text_area(
+#                 "Enter world points as a list of tuples (e.g., [(0, 0), (300, 0), (0, 400), (300, 400)])",
+#                 value="[(0, 0), (300, 0), (0, 400), (300, 400)]",
+#             )
+
+#             if st.button("Confirm Points and Run Homography"):
+#                 try:
+#                     st.session_state["world_pts"] = eval(world_pts_input)
+#                     if len(st.session_state["world_pts"]) != len(st.session_state["img_pts"]):
+#                         st.error("The number of world points must match the number of image points.")
+#                     else:
+#                         st.session_state["homography_ready"] = True
+#                         st.success("Points confirmed! Ready for homography transformation.")
+#                 except Exception as e:
+#                     st.error(f"Error parsing world points: {e}")
+
+# # Perform homography transformation
+# if st.session_state.get("homography_ready"):
+#     st.write("### Running Homography Transformation...")
+#     try:
+#         run_point_est(
+#             st.session_state["world_pts"],
+#             st.session_state["img_pts"],
+#             st.session_state["stitched_image"],
+#         )
+#         st.session_state["homography_ready"] = False  # Reset the flag after execution
+#     except Exception as e:
+#         st.error(f"Error during homography transformation: {e}")

estimate_homography.py

@@ -0,0 +1,119 @@
+import cv2
+import numpy as np
+
+def calculate_homography(in_pts, out_pts):
+    """
+    Calculates the homography matrix H such that in_pts = H * out_pts.
+    :param in_pts: Source points as a numpy array.
+    :param out_pts: Destination points as a numpy array.
+    :return: Homography matrix H.
+    """
+    if isinstance(in_pts, list):
+        in_pts = np.array(in_pts)
+
+    if isinstance(out_pts, list):
+        out_pts = np.array(out_pts)
+
+    mat_A, mat_b = build_sys_equations(in_pts, out_pts)
+    H = np.matmul(np.linalg.pinv(mat_A), mat_b)
+    H = np.reshape(np.hstack((H, 1)), (3, 3))
+    return H
+
+def build_sys_equations(in_pts, out_pts):
+    """
+    Builds the system of equations for homography calculation.
+    :param in_pts: Array of input points.
+    :param out_pts: Array of output points.
+    :return: Matrix A and vector b.
+    """
+    mat_A = np.zeros((np.size(in_pts), 8))
+    mat_b = in_pts.ravel()
+
+    i = 0
+    for x, y in out_pts:
+        # x row
+        mat_A[i][0:3] = [x, y, 1]
+        mat_A[i][-2:] = [-x * mat_b[i], -y * mat_b[i]]
+
+        # y row
+        mat_A[i + 1][-5:] = [x, y, 1, -x * mat_b[i + 1], -y * mat_b[i + 1]]
+
+        i += 2
+
+    return mat_A, mat_b
+
+def fit_image_in_target_space(img_src, img_dst, mask, H, offset=np.array([0, 0, 0])):
+    """
+    Warps img_src into img_dst using the homography matrix H.
+    :param img_src: Source image.
+    :param img_dst: Target image.
+    :param mask: Mask for the destination region.
+    :param H: Homography matrix.
+    :param offset: Offset correction array [x_offset, y_offset, 0].
+    :return: Transformed image.
+    """
+    pts = get_pixel_coord(mask)  # Get all pixel coordinates in the mask region.
+    pts = pts + offset  # Apply offset correction.
+    out_src = np.matmul(H, pts.T)
+    out_src = out_src / out_src[-1, :]  # Normalize to homogenous coordinates.
+
+    out_src = out_src[0:2, :].T  # Extract x, y coordinates.
+    pts = pts[:, 0:2].astype(np.int64)  # Target points in destination image.
+
+    h, w, _ = img_src.shape
+    img_dst = get_pixel_val(img_dst, img_src, pts, out_src, offset)
+    return img_dst
+
+def get_pixel_coord(mask):
+    """
+    Extracts x, y coordinates of white pixels in a binary mask.
+    :param mask: Binary mask.
+    :return: Homogenous coordinates of mask pixels.
+    """
+    y, x = np.where(mask)
+    pts = np.concatenate((x[:, np.newaxis], y[:, np.newaxis], np.ones((x.size, 1))), axis=1)
+    return pts
+
+def get_pixel_val(img_dst, img_src, pts, out_src, offset):
+    """
+    Performs bilinear interpolation to fetch pixel values.
+    :param img_dst: Destination image.
+    :param img_src: Source image.
+    :param pts: Points in the destination image.
+    :param out_src: Corresponding points in the source image.
+    :param offset: Offset correction.
+    :return: Updated destination image.
+    """
+    h, w, _ = img_src.shape
+    tl = np.floor(out_src[:, ::-1]).astype(np.int64)
+    br = np.ceil(out_src[:, ::-1]).astype(np.int64)
+
+    pts = pts - offset[:2]
+    valid_mask = ~np.logical_or.reduce(
+        (np.any(tl < 0, axis=1), np.any(br < 0, axis=1), tl[:, 0] >= h - 1, tl[:, 1] >= w - 1, br[:, 0] >= h - 1, br[:, 1] >= w - 1)
+    )
+    pts = pts[valid_mask]
+    out_src = out_src[valid_mask]
+    tl = tl[valid_mask]
+    br = br[valid_mask]
+
+    tr = np.concatenate((tl[:, 0:1], br[:, 1:2]), axis=1)
+    bl = np.concatenate((br[:, 0:1], tl[:, 1:2]), axis=1)
+
+    weight = np.zeros((out_src.shape[0], 4))
+    weight[:, 0] = np.linalg.norm(tl - out_src[:, ::-1], axis=1)
+    weight[:, 1] = np.linalg.norm(tr - out_src[:, ::-1], axis=1)
+    weight[:, 2] = np.linalg.norm(bl - out_src[:, ::-1], axis=1)
+    weight[:, 3] = np.linalg.norm(br - out_src[:, ::-1], axis=1)
+
+    weight[weight == 0] = 1
+    weight = 1 / weight
+    weight /= np.sum(weight, axis=1, keepdims=True)
+
+    img_dst[pts[:, 1], pts[:, 0], :] = (
+        img_src[tl[:, 0], tl[:, 1], :] * weight[:, 0:1]
+        + img_src[tr[:, 0], tr[:, 1], :] * weight[:, 1:2]
+        + img_src[bl[:, 0], bl[:, 1], :] * weight[:, 2:3]
+        + img_src[br[:, 0], br[:, 1], :] * weight[:, 3:4]
+    )
+    return img_dst

requirements.txt

@@ -0,0 +1,7 @@
+streamlit
+opencv-python-contrib
+opencv-python-headless
+matplotlib
+pillow
+plotly
+streamlit_drawable_canvas