Hardware implementation of a spatio-temporal average filter for real-time denoising of fluoroscopic images

• The paper proposes the FPGA implementation of a spatio-temporal filter for the denoising of fluoroscopic images.
• The filter is able to remove the Poisson noise while preserving edges and motion.
• The circuit has been implemented by using an external DDR2 memory.
• The memory controller and the filter have been optimized to allow real-time processing.
• The circuit is able to process 49 fps if implemented on a StratixIV FPGA by using the 79% of the logic resources.

An electronic system for the real-time denoising of fluoroscopic images is proposed in this paper. Fluoroscopic devices use X-rays to obtain real-time moving images of patients and support many surgical interventions and a variety of diagnostic procedures. In order to avoid risks for the patient, X-ray intensity has to be kept acceptably low during the clinical applications. This implies that fluoroscopic images are corrupted by large quantum noise (Poisson-distributed). Real-time noise reduction can offer a better visual perception to doctors and possible further reductions of the dose.The proposed circuit implements a spatio-temporal filter optimized for the removal of the quantum noise while preserving video edges and the prompt response of the image to the introduction of new features in the field. The filter incorporates information on the dependence of the standard deviation of the noise on the local brightness of the image and performs a conditioned average operation.The proposed circuit is implemented on FPGA (Field Programmable Gate Array) device allowing the real time elaboration of video streams composed by frames with 1024×1024 pixel and uses an external DDR2 (Double Data Rate 2) memory for the storage and the reuse of the fluoroscopic frames needed by the filter.When implemented on StratixIV-GX70 FPGA the circuit is able to process up to 49 fps (frames per second) while using 80% of the logic resources of the FPGA.