Real-time adaptive pixel replacement

Scintillation noise artifacts are a part of intensified imagery for both analog and digital sensors. The high intensity flashes are similar to classic “salt” noise although they often are multiple pixels in extent; they can prove very distracting when utilizing intensified imagery under stressful conditions. In stereo intensified vision system, the fact that artifacts occur at different locations in the left and right sensor increases their ability to distract. Digital intensified sensors are not immune from this problem; however, digital image processing gives us an opportunity to mitigate the problem. A 3×3 median filter is the classic suggested solution to “salt” noise. However, the multiple pixel extent of scintillation noise requires the median neighborhood to be increased to 5×5 for effective suppression. Unfortunately, median also introduces a low pass effect that smoothes the imagery to an unacceptable degree. To overcome this loss of image clarity, we have developed and implemented an adaptive algorithm that is designed to identify scintillation noise. Scintillated pixels are replaced using the 5×5 median while unaffected pixels are left unchanged. The algorithm was tested on a Xilinx XC6SLX150-3 and is capable of operating at a pixel clock of over 220 MHz. With a pixel clock of 140 MHz and a 60 Hz frame rate, the module latency is under 26 μs. We discuss the identification of scintillated pixels and will comparing frames from the raw video, the 5×5 median video and the adaptively replaced 5×5 median.