Consideration of Perceptual Hash using DCT Signs

Abstract

The purpose of this research is to consider a new method to generate a perceptual hash using the sign of the DCT coefficient of the image. DCT is performed by reducing the input image to 32×32 pixels, and the DCT coefficient of the low frequency component 8×8 pixels is extracted by raster scan. A 64-bit hash sequence is generated by setting 1 if the DCT coefficient is 0 or more and 0 if it is negative. The Hamming distance of the hash sequence of different images is 20 or more. On the other hand, it was found that the Hamming distance between images that people feel similar is 10 or less. From many experimental results, it was found that perceptual hash can be generated using the sign of the DCT coefficient of the low frequency component of the image.