Pre-Processing for Image Sequence Visualization Robust to Illumination Variations
Several images (a sequence) may be used to obtain better image quality. This method is perfect for super-resolution algorithms, which improve sub-pixel clarity of the image and allow a more detailed view. It is possible that illumination variations, e.g.
- PDF / 1,569,452 Bytes
- 26 Pages / 439.37 x 666.142 pts Page_size
- 57 Downloads / 196 Views
tract Several images (a sequence) may be used to obtain better image quality. This method is perfect for super-resolution algorithms, which improve sub-pixel clarity of the image and allow a more detailed view. It is possible that illumination variations, e.g. those caused by a light source, lessen the benefits of super-resolution algorithms. The reduction of the quantity of such occurrences by stabilizing variations is important. An enhanced stabilization algorithm is proposed for purposes of reduction of variations in illumination. It is based on the energy contained in wavelet coefficients. In the proposed algorithm, energy plays a role of the memory buffer in memory-based techniques of illumination variation reduction. The benefits of the proposed image stabilization are the higher quality of images and better visualization. Possible applications are in surveillance, product quality control, engine monitoring, corrosion monitoring, micro/nano microscopy, etc. Keywords Illumination variations Parseval relation Energy
Wavelet transform
Super-resolution
I. Kuzmanic´ (&) I. Vujovic´ Faculty of Maritime Studies, University of Split, Zrinsko-Frankopanska 38, 21000 Split, Croatia e-mail: [email protected] I. Vujovic´ e-mail: [email protected] S. M. Beroš J. Šoda Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, University of Split, Rud¯era Boškovic´a bb, 21000 Split, Croatia e-mail: [email protected] J. Šoda e-mail: [email protected]
A. Öchsner et al. (eds.), Design and Analysis of Materials and Engineering Structures, Advanced Structured Materials 32, DOI: 10.1007/978-3-642-32295-2_4, Springer-Verlag Berlin Heidelberg 2013
37
38
I. Kuzmanic´ et al.
1 Introduction Illumination variations generate additive and multiplicative noise, explored in different applications, such as surveillance [1], face recognition [2, 3] or superresolution [4]. These variations can disable higher vision applications, such as motion detection, tracking, super-resolution, pattern/object/action recognition, etc. Therefore, more efficient image sequence visualization requires the suppression of illumination. Super-resolution algorithms are used to obtain higher quality and more details of an image, which is a good starting point for visualization applications. Furthermore, if more details are obtained, more data can be analyzed. High resolution (HR) image is obtained by analyzing inter-pixel motion in low resolution (LR) image sequence in case of super-resolution applications [5]. In such a case, variations in illumination cause false motion influencing the construction of the HR image. Super-resolution can be obtained by different algorithms, either non-wavelet e.g. in [6–9] or wavelet [10–12]. The second generation wavelets are more suited for image super-resolution, because more LR frames lead to irregular sampling [13, 14], and due to the sub-pixel displacement between LR frames. However, assumptions on grid (sampling lattice) structure can be made, although this is not necessary [14]. Although
Data Loading...
