Automatic red-eye detection and correction system for mobile sevices

This paper presents an automatic red-eye detection and correction system. In order to detect red eye, we propose to use two information sources: color around the eyes and eye's round shape. The color information is from the red, highlight, and skin masks. Eye-shape is from our proposed real-time 2-dimensional grouping algorithm called ARTS. Our correction method promises to a natural-looking result. We designed the hardware using Verilog HDL, and successfully built and tested it using an FPGA device, a USB interface board, and a two-megapixel CMOS sensor. With a TSMC 0.25-mum ASIC library, the gate count was 325,167 gates, and the maximum data arrival time was 41.8 [MHz].