Effect of Median Filter on High-level Synthesized Pencil Drawing Style Image Conversion Hardware
Honoka Tani, Akira Yamawaki · 2024
We are developing a Non-Photorealistic Rendering (NPR) library optimized to High-Level Synthesis (HLS) technology, which automatically converts software to hardware.We have created a hardware-oriented software for pencil drawing style image conversion (PENC) which is one of NPRs and built a real-time NPR system on FPGA with a camera and a display equipping our PENC hardware by HLS.This paper tries to reduce a bad effect on the PENC image by inserting a median filter, a noise reduction filter to the camera image, without a performance degradation of the hardware.The experiment evaluates that the effect to the output image caused by where the filter is inserted to the data path, and the effect to the frame rate, circuit size, and power efficiency.As a result, the insertion points of the median filter had no effect on the frame rate, and the developed hardware showed sufficient performance for the software.In addition, depending on the insertion point of the filter, it was possible to reduce the noise in the edges and other areas separately.