Image Compression Based on Direction-Invariant Block Discrete Transform

Based on the original developed models for performing block transforms described in previous own works, the dependences of the results of image comparison in the frequency domain using a two-dimensional cosine basis were obtained. A structure for the algorithm has been proposed, based on the image compression algorithm using DCT, with modifications to the transform. The traditional block transform has been replaced with a modified version according to the described scheme. This further entails a series of changes in the arrangement of the transform results, modifications to the quantization table, and the execution of the zigzag operation. During reconstruction, processing in the frequency domain with elements of invariance to rotations and mirror reflections is possible. Simulation has been conducted on several image databases, yielding comparison results based on the correlation coefficient with templates using a set of frequency components. Software implementing the compression algorithm has been developed, and key performance characteristics have been obtained and compared with the traditional algorithm. The computational complexity did not increase during implementation, confirming the theoretical developments.