SIFT Based Feature Extraction and Matching for Archaeological Artifacts
0 likes79 views
![BIPUL MOHANTO [LION], profile picture](https://cdn.slidesharecdn.com/profile-photo-BIPULMOHANTO-48x48.jpg?cb=1648402975)
The document discusses advanced techniques for feature extraction and matching of archaeological artifacts using methods like SIFT and SURF, which improve upon traditional subjective assessment. It outlines the process of generating feature descriptors for image comparison through techniques like histograms of gradient directions and brute force matching. The research emphasizes enhancing classification accuracy via vector quantization and support vector machines, aiming for a more efficient and reliable identification of artifacts.
SIFT Based Feature Extraction and Matching for Archaeological Artifacts
- 1. SIFT Based Feature Extraction and Matching for Archaeological Artifacts Bipul Mohanto Lecturer, Department of Computer Science and Engineering Varendra University, Rajshahi
- 2. Research Challenge • Traditional way of determining a sculpture’s time-period and geographical area still depending on subjective judgment which is tiring and time consuming. • Experts manually extracts the features of the sculpture’s, then find the best matches with the existing sculptures.
- 3. Image Features • Literally Visual features are not different than image features. • Features: part of information which describes an image, e.g. edges, corners, circles, ellipses, blobs etc. • These can be called a tiny patches which is invariant to image translation, scaling, rotation, illumination differences, or other image parameters..
- 4. Local Feature Detection and Description • Scale Invariant Feature Transform (SIFT) • Dense Scale Invariant Feature Transform (DSIFT) • Speeded Up Robust Feature (SURF) • Oriented FAST and Rotated BRIEF (ORB) • BOW, HOG, RVM, CRF etc.
- 5. SIFT (D. Lowe, University of British Columbia, 2004) 1. Create a scale space of images • Construct a set of progressively Gaussian blurred images • Take differences to get Difference of Gaussian (DoG) pyramid 2. Find local extrema (maxima and minima) in this scale space. Choose key-points from the extrema. 3. For each key-points, in 16×16 window, find Histogram of Gradient (HoG) directions. 4. Create a feature vector out of these.
- 6. 1. Create a scale space of images
- 7. 2. Find local extrema
- 9. Feature Descriptors and Brute Force Matching • Each of the features generates an array of numbers, called feature descriptor. • Brute Force Matching takes the first descriptor from the first image and check with all the descriptors of the second image, returns best result with Hamming Norm. • We can choose user defined matching calculated from Hamming Norm which is actually an array.
- 11. Conclusion and Continue of Work • SIFT features has high dimension in the feature space. • Grouping descriptor to the set of clusters (vocabulary) with vector quantization algorithm using K-means. • Construction of a bag of features (BOF), which calculates the number of features that are entered on each cluster. • Classification (SVM), training bag of features as feature vectors, and determine category of the image.