Cost-effective VLSI Design of Non Linear Image Processing Filters

Abstract

This paper presents a design methodology suitable for the cost-effective and real-time implementation of nonlinear image processing algorithms. Starting from high-level functional descriptions the proposed optimization flow simplifies the designer's duty to achieve a low complexity and low power realization in CMOS technology (FPGA and/or ASIC) with low accuracy loss for the implemented algorithm. As an application case study the paper describes the design of a system, based on a Retinex-like algorithm, to improve the visual quality of images acquired in bad lighting conditions

ASIP-based reconfigurable architectures for power-efficient and real-time image/video processing

“…The dedicated core is synthesized using the same flow and CMOS technology of the ASIP; it has a logic complexity of 90 Kgates, allows for real-time processing of up to 30 Hz CIF (256 9 256 images up to 46 Hz), and has a similar energy cost per pixel, about 45 nJ/ pixel for typical input frames processing. Compared to such results, the r-ASIP synthesis is satisfactory considering its programmable and configurable architecture, opposed to the dedicated one in [32].…”

“…In [32], a dedicated VHDL implementation of the Retinex filter is proposed. The algorithmic type and optimizations (piecewise linearization of nonlinear operators and bit-true arithmetic) are the same as discussed for the r-ASIP design.…”

“…modifying the dynamic of the illumination and enhancing the image details by reflectance processing. Target applications include correction of images and videos acquired in bad lighting conditions, contrast enhancement, and control of dynamic in logarithmic sensors [23][24][25][26][27][28][29][30][31][32]. Most Retinexlike filters exploit a similar structure, sketched in Fig.…”

“…It is also used for other applications such as frame-rate up-conversion, key frame detection, temporal filtering. As far as Retinex is concerned, it is a general purpose algorithm used for linear and logarithmic image sensors in applications such as image enhancement, gamma correction, luminance estimation [23][24][25][26][27][28][29][30][31][32]. Single-scale and multi-scale Retinex operators, with and without temporal filtering, for monochrome or for colored images, have been proposed in literature.…”

Algorithmic and architectural design for real-time and power-efficient Retinex image/video processing

“…9 a Original image, b filtered with our Retinex, and c filtered with the single-scale Retinex proposed in [14,15] the input values to the G and b operators. Following a methodology similar to the one described in [25], several edge distribution laws have been compared to find the best trade-off between circuit complexity and algorithmic accuracy. The latter has been measured with both a PSNRbased objective criterion and a visual perception subjective one.…”