"95% match….These are the words that the computer displayed in bold letters after we ran our first test. We have a crew of six members, who have been working on the project “FACE DETECTION”. The scope for the project is enormous. Although we have implemented some features of our own in the project like face isolation using probability, rgb and ycbcr colour modes, these ideas are yet to be perfected in practice.
To start with, we took a database of over 150 individuals each having 20 images and constructed image vectors from the matrices obtained. These image vectors consisted of 36000 pixels each and hence formed a subspace of R36000. An orthonormal basis of this subspace was found using the PCA(Principal Component Analysis) and SVD(Singular Value Decomposition Theorem). These vectors were arranged in order of their contribution to the database images. These were called the eigenfaces. It was further found that majority of the features of the face depended upon the initial few eigenfaces.
The test images were reconstructed using these eigenfaces. We also used K-means clustering to group various images of the same person in the database. However some of our endeavors failed during execution some of which were- Edge Detection, Hausdorff Distance Measurement, Circular Object Detection, Binary Image Method."
- Akshay Agrawal, Mukesh Kumar Tripathi, Parth Singh, Pooja Narula, Shashwat Goel, Vibhor Kalra
No comments:
Post a Comment