Scalable intelligent median filter core with adaptive impulse detector
Nanduri Sambamurthy,Maddu Kamaraju
DOI: https://doi.org/10.1007/s10470-024-02261-4
IF: 1.321
2024-02-26
Analog Integrated Circuits and Signal Processing
Abstract:This paper introduces a reconfigurable AI-enabled scalable median filter with an adaptive impulse detector designed for FPGA-based real-time imaging systems. Its primary objective is to address the degradation of image quality caused by mixed impulsive noise during real-time image transmission and reception. Existing median filters often struggle to provide real-time image processing results that meet high standards in terms of both accuracy and speed. This approach effectively suppresses noise in real-time images while preserving essential edge details, which are crucial for the performance of real-time imaging systems. The algorithm introduces a novel technique of replacing noisy pixels with the processed central value within the image filtering window. This ensures fidelity to the original pixel, which is vital for applications such as image filter cores. To handle high noise densities in real-time systems, the methodology employs a scalable sorting approach for median filtering and an impulse detector, ensuring robust noise reduction without excessive computational complexity. The AI-enabled scalable median filter system achieves a significant reduction in dynamic power consumption, realizing an impressive 46% decrease in power consumption and an 82% reduction in area compared to the existing system. This is particularly beneficial for addressing resource and power-aware constraints in real-time systems. Comprehensive performance evaluation, including metrics such as PSNR, MSE, IEF, and SSIM, demonstrates the efficacy of the filter in enhancing image quality, a critical factor for the success of real-time imaging systems.
engineering, electrical & electronic,computer science, hardware & architecture