Variants of block frequency transform for image retrieval

Abstract

An accelerated growth of digital media collections has established the need for development of tools for efficient access and retrieval of visual information. The paper presents different variants of block frequency transform to form an array of features for a quick search. Our approach is to perform block frequency transform with elements of invariance for certain rotation angles. We show three main variants macroblock forming. Two variants have different kinds. Mathematical definitions are presented. Graphic views of different macroblock types for discrete cosine transform are showed. A comparison of three transform variants with rotation at different angles is made and presented. It performed in frequency and spatial domains. The execution time does not increase. This approach can be used in various fields related to data processing, data transfer, etc.