Abstract—Transmission estimation is an important step in underwater image restoration based on dark channel prior. To obtain good visual quality, guided image filter is usually adopted to refine the crude transmission map of underwater images. In this paper, we first develop a simplified guided image filter for underwater image restoration to reduce the computational complexity of transmission estimation while maintaining good visual quality. Subsequently, a low-cost architecture for the simplified guided image filter is proposed to achieve real-time Full-HD (1920 × 1080) underwater image restoration. The proposed architecture is synthesized with TSMC 90nm CMOS technology and the result shows that it can operate at 100MHz and support Full-HD image restoration at a throughput of 30frame/s with 10.5K gate counts and 1.35KB on-chip memory. Compared to previous hardware design of guided image filter, the proposed design uses 11.3% of gate counts and 42.2% of on-chip memory to achieve the same frame rate with almost no visual quality degradation.

Index Terms—underwater image, dark channel prior, transmission estimation, guided image filter

Cite: I-Chang Huang, Cheng-Hao Chang, Mu-Fan Lin, Hung-Hsiang Chen, and Shiann-Rong Kuang, "Hardware Implementation of an Efficient Guided Image Filter for Underwater Image Restoration," International Journal of Signal Processing Systems, Vol. 5, No. 3, pp. 94-99, September 2017. doi: 10.18178/ijsps.5.3.94-99