Computer Vision - Work Summary

A short overview of my work:

  • I worked on the implementation of "Fisher Kernel" for classifying images by training the GMM on large dataset. The implementation divided into the following steps, assuming a set of images for each image class:
    1. Generate sifts descriptors for each image. This set is generated for each image in all the classes and called - feaset - descriptor of the image.
    2. Train the GMM classifier on all the generated sifts.
    3. Generate a Fisher Vector using the GMM.

----—For more information see Improving the Fisher Kernel for Large-Scale Image Classification

  • I read an article about SCC-Spectral Curvature Clustering and an overview on its implementation. This article deals with improving the performance of multi-way spectral clustering. The article suggests some nice application using this kind of clustering:
    1. Motion Segmentation under Affine Camera Models
    2. Face Clustering under Varying Lighting Condition
    3. Temporal Segmentation of Video Sequences

----—For more information see Spectral Curvature Clusterin

Throughout the work in this course I was exposed to machine learning, Pattern Recognition ,data mining and some interesting algorithms of Computer Vision. In addition, I got familiar with the research world in general, and with the Computer Vision field, in particular.