Abstract— This paper deals with a major problem encountered in the area of remote sensing consisting of the registration of multi-sensor images. A novel approach for feature matching of multi-sensor satellite imagery is proposed. The feature points are extracted using an improved version of the Harris Corner Detector (IHCD) and are matched using multi- objective optimization technique which incorporates an angle condition and distance condition in the multi-objective fitness function to match corresponding corner-points between the reference image and the sensed image. The matched points obtained in this way are used to align the sensed image with a reference image by applying an affine transformation. The performance of the image registration is evaluated using feature matching accuracy and RMSE and compared with the existing method multi-objective Discrete Particle Swarm Optimization (DPSO). From the performed experiments it can be concluded that the proposed approach is an accurate method for registration of multi-sensor satellite imagery. Index Terms—Improved Discrete Particle Swarm Optimization (IDPSO), multi-objective optimization, multi-sensor image registration. I. INTRODUCTION MAGE registration is a classical problem encountered in many image processing applications where it is necessary to perform geometric alignment of two or more images of the same scene taken by different sensors, at different times, and/or from different viewpoints. In geometric alignment of two images—the reference image and the sensed image, the reference image is the image with respect to which the alignment is carried out and the image which is to be aligned is called the sensed image [1-3]. The registered image is the transformed sensed image which aligns with the reference. The key idea behind the any image registration process is that the sensed image undergoes the registration process and This paper is submitted for review on 19 th October 2016. S. R. Panchal is a research scholar at CHARUSAT, Changa as well as Asst. Professor in the Department of Electronics and Communication Engg., Sardar Vallabhbhai Patel Institute of Technology, Vasad-388306, India. (e- mail: er_sandip79@yahoo.com ). J. K. Bhalani is with the Department of Electronics and Communication Engg., Babaria Institute of Technology, Varnama-391240, Gujarat, India. (e- mail: jaymin188@yahoo.com) A. S. Shete is with the Department of Electronics and Communication Engg., Sardar Vallabhbhai Patel Institute of Technology, Vasad-388306, India. (e-mail: sheteami92@gmail.com) its pixel coordinates are converted into the reference image pixel coordinates. Thus we get the transformed sensed image. The obtained transformed sensed image is then super imposed on the reference image in visually probable way. After completion of this process, we have a larger 2D view of the scene or single output image that is highly informative [2], [3]. In remote sensing applications, some of the challenges encountered in multi-sensor image registration include: (I) differences in pixel intensity and contrast of corresponding regions captured by different sensors; (II) differences in scale that lead to multiple intensity values in one image mapping to a single intensity value in another image; (III) cloud pixels and noise in the images etc [3], [7]. Image registration process can itself generate noise in the registered image, which can be perceived as blurring effect, change in brightness and contrast levels etc. Hence in this process, accuracy can be justified if registered image is devoid of noise [8]. Image registration method is classified into two broad categories, namely area-based registration [1] and feature- based registration [4]. The area based methods are used when pixel intensity provides important and distinctive information [1]. They use some statistical information to measure the degree of similarity of the whole image [5], [8]. The feature based methods are used when image features like edge, point, contours and corners points provide important information [6]. A satellite image with high-resolution can be several hundred megapixels in size and may occupy various spectral bands. Even though detailed information is provided by high- resolution images, it is not efficient to process the whole image because of limited resources such as storage and memory. It also contains local distortions due to different sensors having different angles, paths, and terrain relief, i.e., accuracy is affected by the distribution quality and the number of feature points [10]. For multimodal images, different pixel information is shared by different sensors. It is difficult to find a common region especially when it has different pixel intensity distribution. So it is difficult to match features also [3], [8]. Feature based methods are extensively applied in remote sensing application as compared to area based methods due to their advantages. In area based methods, correspondences is found in the image space whereas in feature based methods correspondences is found in the feature space that represents Multi-objective optimization of satellite image registration using Improved Discrete Particle Swarm Optimization S. R. Panchal, J. K. Bhalani, A. S. Shete I International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 10, October 2016 602 https://sites.google.com/site/ijcsis/ ISSN 1947-5500