Robust and Efficient Image Alignment Based on Relative Gradient Matching

Abstract

Abstract-In this paper, we present a robust image alignment algorithm based on matching of relative gradient maps. This algorithm consists of two stages; namely, a learning-based approximate pattern search and an iterative energy-minimization procedure for matching relative image gradient. The first stage finds some candidate poses of the pattern from the image through a fast nearestneighbor search of the best match of the relative gradient features computed from training database of feature vectors, which are obtained from the synthesis of the geometrically transformed template image with the transformation parameters uniformly sampled from a given transformation parameter space. Subsequently, the candidate poses are further verified and refined by matching the relative gradient images through an iterative energy-minimization procedure. This approach based on the matching of relative gradients is robust against nonuniform illumination variations. Experimental results on both simulated and real images are shown to demonstrate superior efficiency and robustness of the proposed algorithm over the conventional normalized correlation method.Index Terms-Energy minimization, illumination variations, image alignment, image matching, industrial inspection, nearestneighbor search, robust image matching.

Image Matching with Spatially Variant Contrast and Offset: A Quadratic Programming Approach

“…First attempts were done with various correlation-based methods [1] assuming the constant contrast and offset (p = 0). More recently, in response to the failure of traditional correlation techniques to handle the task [4], methods based on robust statistics and low-order polynomial models of contrast and offset such as proposed by Lai [8] made a step forward and branched out into a family of related algorithms [3,15,14]. In these algorithms, the traditional squared-error kernel is replaced by a statistical M-estimator being more robust in the presence of large signal differences and contrast and offset deviations are modelled with a low-order polynomial of pixel x-and y-coordinates.…”

“…One approach is to avoid dealing with the entire N -dimensional space of images and resort to some selective heuristics-based sampling of the pixel space [14]. Unfortunately, this method relies on manually crafting a heuristic requiring human expert intervention.…”

An Improved Adaptive Genetic Algorithm for Image Segmentation and Vision Alignment Used in Microelectronic Bonding

“…The template matching method based on normalized grayscale correlation has the advantages of strong robustness, high positioning accuracy and reliability. However, it can not satisfy the requirements of high speed and high precision packaging due to the inherent contradictions of the algorithm between the speed and accuracy [26]. The feature-based template matching method utilizes certain image features, thus reducing the amount of calculation, so it shows great potential in the applications of microelectronic bonding [27].…”

Quadratic programming vs. concurrent correlation matching under non-uniform image contrast and offset

“…A polynomial model of non-uniform contrast introduced by Lai [3] stimulated a number of subsequent efforts [4,5,6]. Instead of the more conventional but not robust least squares estimator, Lai's matching uses the robust M-estimator [7] and thus a computationally complex numerical gradient search for the best match in the parameter space of both geometric transformation and polynomial coefficients.…”

“…Instead of the more conventional but not robust least squares estimator, Lai's matching uses the robust M-estimator [7] and thus a computationally complex numerical gradient search for the best match in the parameter space of both geometric transformation and polynomial coefficients. This high dimensional space is 978-1-4244-2582-2/08/$25.00 c 2008 IEEE difficult to explore extensively, so various simplifications were made such as a heuristic feature sampling with manual changes of selection parameters for different recognition tasks [6] or matching only contour images after edge detection [5]. Thus the approach was practical with only a small number of polynomial coefficients and failed with non-smooth contrast changes and even shadows [3].…”