UWICL: A Multi-Layered Parallel Image Computing Library for Single-Chip Multiprocessor-based Time-Critical Systems

Architectures and visual-processing applications of multimedia DSPs

“…Vol. 9, 416 -422 (1998)Ideally, it should be possible to complete several operations on the data without writing the intermediate results to external memory [21]. For example, for the resizing example illustrated in Figure 6, it should be possible to complete the vertical and horizontal resizing operations without having to perform multiple read/write memory operations.…”

High-performance image computing with modern microprocessors

“…In addition, the lack of code optimization tools was another hurdle that prevented from efficient code scheduling and data flow debugging. Even with these limitations, the imaging functions carefully programmed in assembly language were able to deliver high performance, e.g., 19 ms for 2D convolution, 14 ms for affine warping, and 75 ms for 2D FFT on a 512 ϫ 512 8-bit image [9,30,31]. Prior to the introduction of the TMS320C80, this kind of image processing execution time has been only possible with dedicated hardwired boards or board sets, e.g., commercial convolver, warper, and special FFT boards.…”

“…Most important, it takes a significant amount of time and effort to develop a high-performance program. In the case of developing an optimized image computing library for the TMS320C80 [30], a typical imaging function, e.g., convolution and wavelets, took about 5 human-months each, while it took 18 human-months for both affine and perspective warping to obtain the best possible performance from the TMS320C80. This is mainly due to the complexity of the TMS320C80 architecture, resulting in a steep learning curve.…”

“…Intel's image processing library for MMX [33], Numerixs image processing library for the TMS320C6x [34], Sun Microsystem's mediaLib [35], and University of Washington Image Computing Library (UWICL) for the TMS320C80 [30] are several examples of imaging libraries. These types of library can provide a well-founded infrastructure of low-level image-computing routines so that higher-level algorithms and applications can be developed and quickly integrated on target systems with minimal custom low-level programming by using the library routines as building blocks.…”

PC-based machine vision system for real-time computer-aided potato inspection

“…The two main reasons for selecting the MS5000 in this application are that: (1) it has been successfully applied to other computation-intensive multimedia and medical imaging applications [8], and (2) it is commercially available along with an imaging library containing many high-performance image-processing routines [9].…”