<title>Impact of multi-scale retinex computation on performance of segmentation algorithms</title>

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

“…Narrow band, in turn, implies a wide impulse filter response and is the main reason for the high complexity of most known algorithms [23][24][25], which are more suited for off-line image processing than for in-line real-time applications. Recent works in literature [30,31] aim at reducing the computational filter complexity to allow real-time and low-power Retinex processing implementation.…”

“…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

“…However, the main constraint for the illumination estimator is that the output signal should be very smooth in the uniformly illuminated parts, and consequently a very large impulse response filter, i.e., a narrow band filter, is mandatory. For the realization of narrow band low-pass filters, different solutions can be found in the literature [3][4][5][6][7][8]: the filter can use a large support and consequently a large number of taps; alternatively, the narrow band can be obtained through several iterations of a small support filter on the same image; a multi-scale solution could also be quite suitable. Unfortunately, all these solutions require a significant computational effort and consequently a real-time and low-power hardware implementation of the system would be problematic.…”

“…The Retinex image filters are based on Land's model for human vision's color perception [1,2]. Several works in the literature extended the basic Retinex-like processing with additional features like the use of multiple spatial scales, color restoration, and dynamic range compression to name but a few [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20]. Many potential applications could benefit from the use of Retinex-based image/video processing, such as vision systems for automotive or avionics safety, video surveillance, medical imaging (e.g., radiography), and consumer electronics (e.g., photo and video cameras and smart phones).…”