An image processing module to recognize style of users based on his images/ selfies.
Current approach:
- Run skeleton identification on each image
- Put the skeleton csv (bodyPart,x1,x2,y1,y2) in a text file with the image in dataset (Code not committed)
- Calculate dense SIFT features for selected bounding box for each image
- Store the BOW histogram of the SIFT features for each image as a vector (This vector has a fixed size unlike SIFT keypoints)
- Index the vectors obtained in above step into a KD tree
- For the query image, compute vector using 1-4 and query KD tree with the obtained vector
Note: This code is V0 and under developement