DCT helps to isolate the image into parts of differing importance based on the image's visual quality. The DCT is similar to the Discrete Fourier Transform since it transforms a signal or image from the spatial domain to the frequency representation. As known in an image, most of the energy is concentrated in the lower frequencies, if we transform an image into its frequency components and neglect the high frequency coefficients, we can reduce the amount of data needed to describe the image, without sacrificing too much image quality.

HOG was firstly proposed by Dalal in 2005, and successfully applied and tested on several variants, with a different type of spatial organization, gradient computation and normalization methods. The HOG features are invariant to geometric and photometric transformations. The implementation of HOG can be achieved by dividing the image into small connected regions, called cells, and for each cell compiling a histogram of gradient directions or edge orientations for the pixels within the cell. The combination of these histograms represents the HOG descriptor. The HOG descriptor is characterized by two main parameters. The first parameter is the size of cells per row and per column. The size of the cell represents the size of the patch involved in the calculation of the histogram. The second parameter corresponds to the number of bins orientation which is mainly used to construct the intervals of the angles of the gradient.

Dear Yousif,

I invite you to see this article:

Article DCT and HOG Feature Sets Combined with BPNN for Efficient Fa...

Best regards,

Pr. Hambaba