Model-Based Mapping of Image Registration Applications onto Configurable Hardware

Abstract

This paper develops techniques for mapping rigid image registration applications onto configurable hardware. Image registration is a computationally intensive domain that places stringent requirements on performance and memory management efficiency. Building on the framework of homogeneous parameterized dataflow, which provides an effective formal model for design and analysis of hardware and software for signal processing applications, we develop novel methods for representing and exploring the hardware design space when mapping image registration algorithms into configurable hardware. Our techniques result in an efficient framework for trading off performance and configurable hardware resource usage based on the constraints of a given registration application.

Functional DIF for Rapid Prototyping

“…As a result, dataflow languages are increasingly popular. Their diversity, portability, and intuitive appeal have extended them to many application areas (e.g., see [4,9,22]). …”

Model-based mapping of a nonrigid image registration algorithm to heterogeneous architectures

“…In order to accommodate the non-linear part of the coordinate transform, the actor connections have to be changed and new actors are introduced. As shown in Figure 3, instead of an input token being an individual coordinate, as in [13], we change it to represent the index of the sub-volume being processed. Here, by a "token" we mean a single datum that is transferred along a dataflow graph edge.…”

“…We base our dataflow model for non-rigid registration on the dataflow model for the rigid registration as presented by Hemaraj et al in [13] (shown in Figure 2). Our dataflow model of the complete non-rigid registration system retains the remaining parts of the dataflow model for rigid registration as we separate the nonrigid algorithm into a linear part (global matrix multiplication) and a nonlinear part (B-splines interpolation).…”

“…In this work, we leverage on the dataflow interchange format (DIF) [12], which provides integrated support for a variety of dataflow semantics and features, including dynamic and reconfigurable dataflow behavior. In [13], Hemaraj et al proposed a new approach using dataflow modeling techniques to capture concurrency and dependencies for accelerating the calculation of similarity between two images using mutual-information. It is critical to analyze this calculation carefully because it is the most computationally intensive component in mutualinformation-based image registration.…”

Model-based mapping of reconfigurable image registration on FPGA platforms

“…We present area and power calculations that characterize our proposed architectures. In terms of modeling methodology, we present useful refinements of our previous work on developing and applying novel data flow-based models and analysis methods of image registration applications [8]. These methods provide a framework for mapping and optimizing these applications onto embedded architectures.…”