, pp.34-38 http://dx.doi.org/10.14257/astl.2014.45.07 Sensory Fusion for Image Sungjun Park, Wansik Yun, and Gwanggil Jeon 1 Department of Embedded Systems Engineering, Incheon National University, 119 Academy-ro, Yeonsu-gu, Incheon 406-772, Korea gjeon@incheon.ac.kr Abstract. In this paper, we propose a near infrared images upsampling method using RGB camera information. In general, the resolution of near infrared images is lower than that of RGB images by factor of 4. Simulation results show the visual performance. It is obvious that the details of result image are clearer than that of low-resolution image. Keywords: Sensor, image fusion, near infrared image. 1 Introduction There are many different kinds of images such as RGB images acquired by CCD camera, near infrared (NIR) images acquired by NIR camera, and panchromatic multispectral image acquired by satellites. Although those image are considered to be transferred with the best quality with highest resolution, however due to some issues, those data are conveyed with low resolution. Normally, the resolution is downsampled with factor of 4 or more, therefore at the receiver side, receivers need additional information to reconstruct the low resolution images. This technique is called image fusion. Currently the concept of sensor fusion is very broad and the fusion can take place at the signal, pixel, feature, and symbol level. More details are introduced in [1]. In this paper, we propose a new method to enhance the low resolution NIR images to high resolution NIR images. Here, we use high resolution RGB image information. The rest of this paper is composed as follows. In Section 2, we present the proposed method. In Section 3, experimental results are obtained to show the feasibility of the proposed design. Finally, Section 4 presents our conclusions. 2 Multi-sensor Image Fusion Image sensor fusion is the merging of sensory data or data obtained from sensory data from different sources. After image sensor fusion, we expect the images become more accurate and more complete. In computer vision, multi-sensor image fusion is the process of merging related information from two or more images into a single image. Therefore, the resulting image becomes more informative than any of the input ISSN: 2287-1233 ASTL Copyright 2014 SERSC
images. In this paper, we propose a system where low resolution NIR image and high resolution RGB image is used as input images. As NIR images have lower resolution than that of RGB images, the resulting image of image fusion may be more informative than both images. Here, we use training library which is obatiend by number of test images, such as 25 RGB and NIR images. Fig. 1. Learning-based super resolution algorithm. Copyright 2014 SERSC 35
3 Result Images (a) (b) (c) Fig. 2. Test image #20: (a) Original low-resolution NIR image, (b) Original high-resolution RGB image, (c) multi-sensor method applied high-resolution NIR image. Simulation was conducted on 25 RGB and NIR images given by Sadeghipoor et al. Figures 2 and 3 show the original low-resolution NIR image, original high-resolution RGB image, and its image fusion results. As it is can be seen in both figures, the results NIR images have higher resolution than its previous NIR images, and its details (edge areas) are seen obvious. 36 Copyright 2014 SERSC
4 Conclusion In this paper, we propose a new near infrared images upsampling approach using RGB information. Experimental results show the proposed system gives better performance than the low resolution images. (a) (b) (c) Fig. 3. Test image #25: (a) Original low-resolution NIR image, (b) Original highresolution RGB image, (c) multi-sensor method applied high-resolution NIR image. Copyright 2014 SERSC 37
Acknowledgment. This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Science, ICT and Future Planning(2013R1A1A1010797) References 1. H. Li, B.S. Manjunath, and S.K. Mitra, Multi-sensor image fusion using the wavelet transform, IEEE ICIP1994, vol. 1, 1994, pp. 51-55. 38 Copyright 2014 SERSC