An efficient global registration method for multi-exposure images

Page 1

Volume 2, Spl. Issue 2 (2015)

e-ISSN: 1694-2310 | p-ISSN: 1694-2426

An efficient global registration method for multi-exposure images Harbinder Singh, Vinay Bhatia Electronics and Communication Engineering Department, Baddi University of Emerging Sciences and Technology Baddi, Solan 173215, India e-mail: harbinder.ece@baddiuniv.ac.in, vinay.bhatia@baddiuniv.ac.in Abstract— In this paper we present a method for eliminating global misalignments between a sequence of images captured at different exposures with horizontally moving camera. The proposed method utilizes template matching based on normalized correlation to find the correspondences between different input images. The best known correspondence is used to transform a set of images to a single coordinate system and eliminate any global misalignment (including translation). The proposed registration technique works well for under-and over-exposed images usually required for High Dynamic Range (HDR) image generation and construction of HDR panoramic image. Any camera movement results blurry HDR image. The results show that the image registration accuracy of our method is same when images captured at large exposure difference. Uses of different 2-D geometric coordinate transformations in consecutive images are also discussed. Keywords — Image registration, Multi-exposure, Template matching, High Dynamic Range Image

I.

INTRODUCTION

Image registration of two-dimensional images is a fundamental process in digital image processing to estimate the underlying correspondence between two or more images. Image registration process compensates misalignments caused by any movement of camera (global motion) or dynamic object in a scene (local motion). The variety of applications worthwhile: HDR image generation, image recognition, image mosaics, artifact reproduction, image fusion and many others such as augmented reality in graphics and map building in robotics. Image registration methods can be divided into feature and area based methods. Similarity measures are used in image registration process to determine correspondence between all points in two or more 2-D images [1] and [2]. In the case of HDR image generation HDR radiance maps are recovered from the differently exposed images [3]. Most techniques assume that the input images captured at different exposure values with a fixed camera. One of the main requirement of compositing HDR image from multiple exposures using the technique proposed by Debvec and Malik [3] is that the camera be absolutely still during image

291

capturing process. Unfortunately, even tripod mounting will sometimes allow slight shift in the camera. The proposed method is dedicated to problem of global registration (including translation) of images captured by hand- held camera at varying exposure values. Method is based on template matching of reference image with windows of same size in other images, which are captured for different EV. The similarity metric used here is normalized correlation to identify window that is most similar to the template of reference image. In the proposed image alignment technique normalized correlation is used for the estimation of translation parameters. Different similarity metrics have been developed and the selection of similarity metric is depends on types of images provided for alignment. A review of various similarity metrics used in image alignment is given in [1]. Our method is fully automatic and compensates global translation of multiexposure images. A review of previous work related to image registration and HDR image generation is given in Section 2. In Section 3, we describe our method for image alignment, and measure normalized correlation of several images captured at different exposure values. Various geometric transformations to modify the spatial relationships between pixels in input and output images are also discussed in Section 3. In Section 4, we discuss the implementation of algorithm, and finally, Section 5 concludes the paper. II.

RELATED WORK

In recent years many algorithms have been proposed to capture HDR images from bracketed images (standard LDR images) [3], [4], [5], and [6].The researchers suggest using tripod to avoid any global movement. The problem of image alignment of multi-exposure images to capture HDR image was proposed in [7] and [8].They were employing conversion of input photographs in to percentile threshold bitmaps for image alignment. We have found out that few conventional approaches dicussed in [1] and [2] to image alignment usually fail when applied to images with variable exposures. As shown in Fig. 1 edge detection filters are exposure dependent, where edges appear and disappear at different exposure levels. Therefore

BUEST, Baddi

RIEECE -2015


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.