Kahinee - A spirited story teller

[Report]
Various algorithms like Local Binary Patterns(LBP), Constrained Local Models (CLM) regularized by Land-mark Mean Shift and Viola-Jones along with CAMSHIFT and KLT were implemented. The results of these algorithms were compared and considering the strengths and weaknesses of each of them a hybrid approach using Viola-Jones and CAMSHIFT was developed for real time face detection.

Protein Classification

We developed a Support Vector Machine based protein classifier. The binary classifier was trained to detect whether the given protein sequence is a Cyclin protein or not. The classifier was trained using dipeptide features and linear, radial and sigmoid kernels. Implemented 5-fold cross-validation and compared the results.

Image reference: https://en.wikipedia.org/wiki/Cyclin_E1

JPEG Compression

[Report]
A compression model takes the raw input data and represents it in such a form that the space required to store the modified data is less than the original data. This project presented simulation of one such technique called JPEG compression. It exploits the fact that human eye is less sensitive to high frequency variation and color variation in an image. Hence, we obtain a compressed image by removing such information. The quality of the compressed image can be varied by changing the Q-factor. Several techniques like DCT, iDCT, quantization, zigzag traversal, run length encoding and decoding were implemented to simulate the model.