Course Highlights:

This course highlights the issues and solutions for the real time implementation of Digital Image Processing systems on embedded platforms based on the following aspects like:

· Images of any reasonable size will not fit in on-chip RAM, which has just been identified as a rate-limiting factor in algorithm performance. Because the latencies involved with accessing off-chip RAM are so severe, an optimal 2D DWT transform should incorporate strategies to circumvent this problem.
· As it turns out, in the fully optimized case one should use Direct Memory Access (DMA) to increase the speed of the block memory copy, which has the added benefit of freeing the processor for other duties.
· Developing fixed-point algorithm implementations, where any floating-point calculations are carried out using integer representations of numbers and the decimal point is managed by the programmer, opens up a whole new can of worms. The algorithm developer must now contend with a slew of new issues, such as proper data scaling, quantization effects of filter coefficients, saturation and overflow, to name just a few.
· And even still, there are many other issues that the preceding discussion omits, for example the inclusion of assembly language, vendor-specific compiler intrinsics, and a real-time operating system, but the general idea should now be clear. Even just a few of these issues can turn a straightforward image processing algorithm into a fairly complicated implementation. Taken together and all at once, these concerns may seem overwhelming to the uninitiated.

Course Objectives

Developing embedded Digital Image Processing (DSP) applications is a complex task influenced by many parameters. But by using a systematic approach the result can be an efficient implementation. Because the main objective of this course are efficient implementations of image processing algorithms running on embedded DSP systems, it is not sufficient to simply present an algorithm and describe a first-cut implementation that merely produces the correct output. The primary goal is efficient algorithm implementations, while a secondary goal is to learn how to utilize the appropriate TI technologies that aid in the development and debugging of code that can be used in real-world applications.

The transition from theory to implementation is a jump that deserves more attention than is typically given. In particular, taking a description of an image processing algorithm and coding an efficient implementation in the C language on an embedded fixed-point and resource-constrained DSP is arduous and is mainly focused in this course.

The main content in this course contains introductory material and is important background information on the various tools employed throughout the course.
1. Image Enhancement via spatial processing techniques (point-processing).
2. Image Filtering (linear, non-linear, and adaptive).
3. Image Analysis (edge-detection and segmentation).
4. Wavelets (with applications to edge detection and image enhancement).

Due to the challenging nature of embedded development, the strategy is to start off simple and then progressively delve deeper and deeper into the intricate implementation details, with the end purpose always being an efficient algorithm running on the DSP.

Who Should Attend

DSP professionals, graduates and research scholars who would like to know profound aspects real time implementation of Digital Image Processing Systems on DSP processor based embedded system such as:
· Advanced Image Processing Algorithms
· Architecture details for TI's TMS320C67xx
· Fixed Point and Floating Point DSP programming
· Optimization methods and techniques for efficient DSP implementation and performance enhancement
· Real Time DSP System implementation aspects and issues