Midpoint presentation

Here is the slides for midpoint presentation.
http://dl.dropbox.com/u/70433681/MidpointPresentation.pptx

•Progress so far 

        –Gradients computed

           –Got some bugs when accumulating gradients for each block of cells. (Bugs fixed but descriptors not verified)

•Steps to detect logo

–Sample and compute overlapping HOG features for reference image. (Codebook, offsets from center position are also saved)

–Sample and compute overlapping HOG features for target image.

–For each HOG descriptor in target image, find k closest matches and vote corresponding center position.

–Center candidates above threshold are considered as detected logo center.

–The above steps are repeated for different scales.