Handwritten Digits Recognition Using PCA of Histogram of Oriented Gradient

This paper presents a multiclass classifier based on principal component analysis (PCA) of histogram of oriented gradient (HOG) for accurate and fast recognition of handwritten digits. HOG is known as an effective feature descriptor for computer vision and image processing, and PCA has shown its ability for fast multiclass recorgenition. By combining PCA with HOG, the PCA-of-HOG based classifier is developed. The proposed algorithm was applied to the MNIST database of handwritten digits to demonstrate its performance in comparison with classifiers based on PCA of raw input data.