Resolution-Aware Heterogeneous Cpu–gpu Computing for Image Preprocessing

This article presents a CUDA-based image preprocessing approach for heterogeneous CPU–GPU systems, focusing on image-resolution-aware CUDA configuration and efficient GPU memory utilization. The CPU performs host-side data preparation, control, and transfer operations, while image filtering and brightness enhancement are executed on the GPU. CUDA block dimensions are selected according to the tested image resolution, using 8×8, 16×16, and 32×32 configurations. The approach was evaluated using images with resolutions from 512×512 to 3840×2160 pixels. The reported measurements show that the CPU–GPU configuration achieves lower execution times than the compared static CPU and GPU implementations, providing speedups of 2.19× to 3.69× over CPU execution. The results demonstrate the potential of resolution-aware CUDA configuration and heterogeneous CPU–GPU organization for accelerating the tested image-preprocessing pipeline.