Image Registration of Very Large Images via Genetic Programming

📝 Abstract

Image registration (IR) is a fundamental task in image processing for matching two or more images of the same scene taken at different times, from different viewpoints and/or by different sensors. Due to the enormous diversity of IR applications, automatic IR remains a challenging problem to this day. A wide range of techniques has been developed for various data types and problems. However, they might not handle effectively very large images, which give rise usually to more complex transformations, e.g., deformations and various other distortions. In this paper we present a genetic programming (GP)-based approach for IR, which could offer a significant advantage in dealing with very large images, as it does not make any prior assumptions about the transformation model. Thus, by incorporating certain generic building blocks into the proposed GP framework, we hope to realize a large set of specialized transformations that should yield accurate registration of very large images.

💡 Analysis

Image registration (IR) is a fundamental task in image processing for matching two or more images of the same scene taken at different times, from different viewpoints and/or by different sensors. Due to the enormous diversity of IR applications, automatic IR remains a challenging problem to this day. A wide range of techniques has been developed for various data types and problems. However, they might not handle effectively very large images, which give rise usually to more complex transformations, e.g., deformations and various other distortions. In this paper we present a genetic programming (GP)-based approach for IR, which could offer a significant advantage in dealing with very large images, as it does not make any prior assumptions about the transformation model. Thus, by incorporating certain generic building blocks into the proposed GP framework, we hope to realize a large set of specialized transformations that should yield accurate registration of very large images.

📄 Content

Ref: IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
Workshop on Registration of Very Large Images, pp. 323—328, Columbus, OH, June 2014.

1

Abstract

Image registration (IR) is a fundamental task in image processing for matching two or more images of the same scene taken at different times, from different viewpoints and/or by different sensors. Due to the enormous diversity of IR applications, automatic IR remains a challenging problem to this day. A wide range of techniques has been developed for various data types and problems. However, they might not handle effectively very large images, which give rise usually to more complex transformations, e.g., deformations and various other distortions. In this paper we present a genetic programming (GP)- based approach for IR, which could offer a significant advantage in dealing with very large images, as it does not make any prior assumptions about the transformation model. Thus, by incorporating certain generic building blocks into the proposed GP framework, we hope to realize a large set of specialized transformations that should yield accurate registration of very large images.

1. Introduction Image registration (IR) is an important, significant component in many practical problems in diverse fields where multiple data sources are integrated/fused, in order to extract high-level information as to the contents of the given scene. A wide range of registration techniques has been developed over time, where typically, specific domain knowledge is taken into account and certain a priori assumptions are made, e.g., with respect to the transformation model used, specific bounds on its parameter values, etc. In contrast, in this paper we present a genetic programming (GP)-based approach for IR. GP is part of a family of evolutionary algorithms (EAs) which are stochastic optimization methods whose goal is to find an “optimal” solution or a set of solutions with respect to certain objective(s). Exploiting this strength of GP for search and optimization problems [8-9], we use GP and image processing techniques to search efficiently for an “optimal transformation” with respect to a given similarity measure. The novel aspect of our proposed GP algorithm is that it does not make any prior assumptions about the transformation model. Therefore, incorporating various building blocks into the GP framework could yield potentially a large pool of transformations. The advantage of this approach is that it offers, in principle, much greater flexibility in the registration of very large images which give rise to various, relatively complex transformation types. We present good results on some real datasets on small images and compare them to those obtained due to a recent method assuming a simple transformation model. We also present initial results on some real datasets containing larger images on simple transformations. From a pure GP perspective, other transformations, possibly more complex ones, could be searched for and discovered, as long as the GP is supported by the proper building blocks.
   The paper is organized as follows. Section 2 reviews IR components that are essential for our solution. Section 3 presents a brief GP background and provides the motivation for our GP-based IR approach. Section 4 presents an overview of related evolutionary-based solutions for IR.
   Section 5 provides a detailed description of our suggested GP-based approach. In Section 6 we present our initial empirical results. Section 7 makes concluding remarks.

2. Image Registration Image registration involves searching for a transformation that generates a maximal match in the overlap between the reference image and the transformed sensed image. It uses a similarity measure to assess the quality of a specific transformation.
   In this work, we focus on one of the most common measures, mutual information (MI) [14],[18], inspired by information theory [17]. MI is a measure of statistical dependency between two datasets that has been applied in a robust and efficient manner to IR. Analogous to the Kullback-Leibler measure [15] for the distance between two distributions, MI of two images measures the degree of dependence between the gray values in the area of overlap, defined as, (1) (, ) = ( , ) ( , ) ( )( ) ,

Image Registration of Very Large Images via Genetic Programming

Sarit Chicotay Dept. of Computer Science Bar-Ilan University Ramat-Gan, Israel saritc@gmail.com

Eli (Omid) David Dept. of Computer Science Bar-Ilan University Ramat-Gan, Israel mail@elidavid.com
Nathan S. Netanyahu Dept. of Computer Science Bar-Ilan University Ramat-Gan, Israel nathan@cs.biu.ac.il

2 The assumption is that maximal dependence between the gray values of the images achieved when they are correctly aligned, while misregistration results in a decre

This content is AI-processed based on ArXiv data.