HOG feature extraction cell size
27 views (last 30 days)
Show older comments
I use hog feature extraction in images to detect the horizon line. Default cell size for hog feature extraction is 8x8. When I use 16x16 I get better result of the accuracy of identifying horizon line. But in some images 8x8 is working better. However in general 16x16 is more successful. I didn't understand why 8x8 is working better in some images while 16x16 is more successful in general? What is the effectiveness of cell size in hog feature extraction?
Amy on 21 Dec 2017
From the documentation for extractHOGFeatures: To capture large-scale spatial information, increase the cell size. When you increase the cell size, you may lose small-scale detail
There is not one optimal cell size. The best size depends on your data.
Find more on Recognition, Object Detection, and Semantic Segmentation in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!Start Hunting!