研究目的
To develop a simple, parameter-free algorithm for generating HDR images from a single LDR image using highlight removal, and to verify its hardware implementability using HLS tools.
研究成果
The proposed parameter-free algorithm effectively generates HDR images from single LDR images using highlight removal, with improved quality and processing speed compared to existing methods. HLS-based implementation confirms hardware feasibility, achieving high similarity between software and hardware outputs (SSIM ≥98.87%, PSNR ≥39.90 dB). Future work includes complete SoC implementation and further optimizations.
研究不足
The algorithm assumes that highlight areas are not fully saturated and surfaces are inhomogeneous. It may not perform well with fully saturated highlights or homogeneous surfaces. The HLS implementation is optimized for specific FPGA resources and may require adjustments for other hardware. The study focuses on PL side development; full SoC implementation is not completed.
1：Experimental Design and Method Selection:
The study involves designing an algorithm for HDR image generation from a single LDR image using highlight removal based on statistical image information. The methodology includes software simulation in MATLAB and hardware implementation using HLS tools for FPGA.
2：Sample Selection and Data Sources:
Ten test images (Doll, Stone, Hen, Idol, Red Ball, Face, Fish, Bear, Green Pear, Cups) were used for evaluation, selected to cover various scenarios with highlights.
3：List of Experimental Equipment and Materials:
Vivado HLS v2016.4 tool, MATLAB software, Xilinx Zynq device (FPGA), PC with Windows 7 64-bit, Intel Core i7-3770K CPU, 12 GB RAM.
4：4 tool, MATLAB software, Xilinx Zynq device (FPGA), PC with Windows 7 64-bit, Intel Core i7-3770K CPU, 12 GB RAM.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The algorithm was first developed and tested in MATLAB. Then, it was converted to HLS C code, optimized for FPGA implementation using Vivado HLS, with steps including C-synthesis, C-simulation, RTL verification, and IP packaging. Performance was evaluated using no-reference metrics (HB, E, NIQE, CPCQI) and full-reference metrics (SSIM, PSNR).
5：Data Analysis Methods:
Data analysis involved comparing software and hardware outputs using SSIM and PSNR for similarity, and no-reference metrics for image quality assessment. Statistical comparisons were made with state-of-the-art methods.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
