Talk:Local Binary Patterns

From Scholarpedia
Jump to: navigation, search

    This article gives a comprehensive survey about one of the most sucessful feature descriptor (LBP) in computer vision. It clearly demonstrates what LBP is, how it works, what are its major applications and how it can be extended. Overall, I think this is a very good article and strongly recommend it to be accepted.

    my comments: It would be interesting if some discussion on why this simple descriptor is so powerful were added. LBP is simple, efficient and effective in many computer vision applications as demonstrated in the last section. But why? LBP starts by calculating the gradient of local patch, thresholding them at the value of the central pixel, then histogrmming them to make a descriptor. So it will be interesting if the role played by each of the three steps in various applications is discussed. A recent work on the extension of LBP, called Local Ternary Pattern (LTP), shows that carefully controling the thresholding step could improve the robustness of LBP against noise which may caused by various illumination preprocessing algorithms.

    Personal tools

    Focal areas