BTP-2009

Name : Subodh Misra

Roll No. : Y6481

E-Mail : subodhm@iitk.ac.in

Advisor : Prof. Amitabh Mukherjee

Title : Developing Open Source code for Pyramidal Histogram Feature Sets

Summary : Bosch and Zissermann technique uses PHOG and PHOW descriptors for image classification. So I will try to implement PHOG and PHOW features in C++ which has not been done so far. Subsequently I will contribute my code to OpenCV library. So I will be extracting foregroung objects from the input videos and do the classification of all those foreground objects. I will use Caltech101 dataset to build the model for classification and use these foreground objects as test images.

BTP Proposal (Updated 1st October 2009)
BTP Term 1 report (Updated 1st October 2009)
BTP Final Report (Updated 15th April 2010)
PHOW Code
README for PHOW code :
1. Directory Structure : Five directories need to be made
images, reg, sifts, sifts_test, vocab
2. Procedure :
-Put all images in 'images' directory
-Run ./regions to get regions.txt files in 'reg' directory
-Run ./sifts to get SIFT features of all images in 'sifts' directory
-Transfer some test image SIFT features files from 'sifts' to 'sifts_test' directory
-Run ./readfeatures_part1 to get vocab.txt in 'vocab' folder
-Run ./readfeatures_part2 that will print PHOW descriptors

PHOG Code
README for PHOG code :
Usage is included in PHOG_CODE.tar
Dataset