Geometry-Aware Transform Pruning for Omnidirectional Image Compression

Abstract

This study introduces a latitude-adaptive discrete cosine transform pruning method designed for low-complexity omnidirectional image compression. The proposed approach exploits the Nyquist-Shannon sampling theorem together with the geometric characteristics of the de facto sphere-to-plane mapping for omnidirectional media—the equirectangular projection—to selectively apply pruned transforms in block-based codecs. By adapting the transform support according to the spatial redundancy introduced by equirectangular sampling distortions, the method reduces unnecessary transform coefficients while preserving the relevant visual information. The proposed framework decreases the arithmetic complexity of both the transform and quantization stages, achieving substantial reductions in the number of additions and multiplications compared with the conventional JPEG pipeline. Experimental results on benchmark high-resolution omnidirectional images demonstrate that the proposed solution attains compression efficiency comparable to that of state-of-the-art low-complexity block-based techniques in terms of bitrate and image quality. These results highlight the potential of the proposed framework for resource-constrained transform-based omnidirectional image codecs.