A Framework for Scheduling and Mapping of Image Signal Processing Pipeline on Heterogeneous Architectures Applied to Video Enhancement Algorithms

Video contents and applications become integrated in everyday life and demands for high-quality videos, such as 4K, are increasing. Even modern optical equipment introduces inevitable noises that may heavily mask video content. Environment conditions such as low light, results in more video quality degradation. Therefore, a video-enhancement technique is required to reconstruct the original non-noisy video, but such algorithms require extensive processing resources to reach a reasonable (real-time) performance and power consumption. Heterogeneous processors and platforms can provide the required performance and power of a video processing algorithm by efficient design. This requires a suitable algorithm for scheduling the algorithm tasks on the processors of the heterogeneous architecture. The long-term objective in our university-industry collaboration is to propose a target architecture-independent framework, including scheduling and mapping algorithm, for accelerating an image-processing pipeline and apply it to video enhancement such as denoising. Our framework will consist of four main parts: a) modeling, b) scheduling and mapping, c) target-independent video enhancement library and, d) objective evaluation.

Faculty Supervisor:

Aishy Amer

Student:

Kazem Cheshmi

Partner:

TandemLaunch Technologies Inc.

Discipline:

Engineering - computer / electrical

Sector:

Information and communications technologies

University:

Concordia University