Classical CV + ML: SIFT, HOG, SVM

Best for: Smaller/legacy vision systems

How it works

Hand-crafted descriptors encode local appearance before deep features existed. HOG computes image gradients and accumulates their orientations into histograms over spatial cells; SIFT locates scale-space extrema and builds orientation histograms around keypoints, giving a gradient-based descriptor $f=\text{hist}(\nabla x)$. These features feed a classical classifier such as an SVM ($\min\tfrac12\|w\|^2$ s.t. $y_i(w^\top\phi(x_i)+b)\ge 1$). Effective for small or legacy vision systems where data or compute is limited.

Common fields

Industrial inspection · embedded systems