How to compute a descriptor

[chulminy]

https://medium.com/@ibabin/an-overview-of-sift-69a8b42cd5da

A student inquired about the process of computing a descriptor from a histogram.

Following the detection of a feature, a 16 x 16 pixel region is examined and subdivided into 4 x 4 cells, with each cell containing 16 pixels.

For each pixel, a gradient is calculated, and these gradient values are allocated into one of 8 bins.

Consequently, each cell is characterized by a histogram consisting of 8 bin values. Therefore, for the 4 x 4 cells, there are a total of 128 values (8 bin x 4 x 4 cells).

Prior to comparison, these values undergo normalization to adjust for variations in intensity between different images.