Course Name: Computer Vision Lab Course Code: CSP-422

Experiment:1.3

Aim:
Write a program to analyze the impact of refining feature detection for image segmentation.

Software Required: Google Colab

Description:
Refining feature detection for image segmentation involves improving the quality and accuracy of detected
features. Techniques for this purpose include:

1. **Multi-scale Analysis**: Detect features at different scales using algorithms like SIFT or SURF, capturing
objects of varying sizes.
2. **Non-Maximum Suppression**: Eliminate less significant feature responses produced by noise or
variations, retaining the most salient features.
3. **Adaptive Thresholding**: Dynamically adjust detection thresholds based on local image statistics for
varying lighting or contrast conditions.
4. **Edge Refinement**: Enhance detected edges using methods like edge thinning, linking, or contour
enhancement.
5. **Feature Filtering and Selection**: Remove irrelevant or false features, considering quality measures,
spatial distribution, or context.
6. **Feature Fusion**: Combine different feature types or descriptors to create a more comprehensive
representation of image content.
7. **Contextual Information**: Incorporate relationships between neighboring pixels or features, utilizing
spatial constraints, semantic priors, or contextual cues.
8. **Deep Learning-based Approaches**: Train deep learning models like CNNs to directly detect and
refine features, capturing complex patterns and context for improved accuracy.

Name: Ayushman Kumar UID: 20BCS8232

 Course Name: Computer Vision Lab Course Code: CSP-422

Pseudo code/Algorithms/Flowchart/Steps:

• Import OpenCV as cv2.

• Define image paths for the test image and dataset.

• Read the test image and create an ORB feature detector.

• Detect key points and compute descriptors for the test image.

• Initialize an empty list called store for matching results.

• Loop through dataset images, detecting key points, computing descriptors, and
calculating matching scores.

• Sort matching results by score, from highest to lowest.

• Print results and display images with their scores.

• Wait for a key press to close displayed image windows.

• Close all image windows when the program finishes.

Implementation:
import cv2
test_image = 'imagetest.png'
dataset = ['test1.jpg', 'test2.jpg', 'test3.jpeg']

testing_image = cv2.imread(test_image)

orb = cv2.ORB_create()
kp_target, des_target = orb.detectAndCompute(testing_image, None)

store = []

for image_path in dataset:

dataset_image = cv2.imread(image_path)
kp_dataset, des_dataset = orb.detectAndCompute(dataset_image, None)

Name: Ayushman Kumar UID: 20BCS8232

 Course Name: Computer Vision Lab Course Code: CSP-422

bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des_target, des_dataset)

matches = sorted(matches, key=lambda x: x.distance)

good_matches = [m for m in matches if m.distance < 75]

matching_score = len(good_matches) / len(kp_target)

store.append((image_path, matching_score))

store.sort(key=lambda x: x[1], reverse=True)

for result in store:

print(f"{result[0]} - Matching Score: {result[1]:.2f}")

for result in store:

image_path = result[0]
matching_score = result[1]
img = cv2.imread(image_path)
cv2.imshow(f"Matching Score: {matching_score:.2f}", img)
cv2.waitKey(0)

Images used-
Data Set images-
Test Image-

Name: Ayushman Kumar UID: 20BCS8232

 Course Name: Computer Vision Lab Course Code: CSP-422

Output:

Name: Ayushman Kumar UID: 20BCS8232