LUM Smoother with Smooth Control for Noisy Image Sequences
- PDF / 2,073,908 Bytes
- 11 Pages / 600 x 792 pts Page_size
- 99 Downloads / 174 Views
UM Smoother with Smooth Control for Noisy Image Sequences Rastislav Luká Department of Electronics and Multimedia Communications, Technical University of Košice, Park Komenského 13, 041 20 Košice, Slovak Republic Email: [email protected]
Stanislav Marchevský Department of Electronics and Multimedia Communications, Technical University of Košice, Park Komenského 13, 041 20 Košice, Slovak Republic Email: [email protected] Received 7 December 2000 and in revised form 3 May 2001 This paper focuses on adaptive structure of LUM (lower-upper-middle) smoothers for noisy image sequences. For the balance between noise suppression and signal-detail preservation, the LUM smoothers are widely used in smoothing applications. The amount of smoothing done by LUM smoothers is controlled by tuning parameter. However, the smoothing level is fixed for whole image. Thus, the excessive or insufficient smoothing can be performed. This problem is solved by a new method based on the adaptive controlled level of smoothing. A new method has excellent performance of the noise reduction in the environments corrupted by the impulse noise. In addition, minimal signal-detail and motion blurring can be observed. The performance of proposed method is evaluated through objective criteria and compared with traditional temporal, spatial, and spatiotemporal LUM smoothers. Keywords and phrases: image sequence processing, impulse noise, LUM smoother, adaptive control.
1. INTRODUCTION The processing of image sequences is widely used in medical imaging, computer and robot vision, video communications (e.g., phone video conferencing) and television transmission chain. However, the time-varying images or image sequences can be considered as spatiotemporal data [1], that is, a time sequence of two-dimensional (2D) images. The fact that the third dimension, that is, time is included increases computing complexity and time processing. In practice, image signals interfere with impulse noise included by atmospheric noise, such as lightning spikes and spurious radio emission in radio communication, and relay switching noise in telephone channels. In addition to these natural non-Gaussian noise sources, there is a great variety of man-made sources such as electronic devices. It is evident that for sufficient restoration of degraded image signals some filtering techniques must be used. Therefore, few methods of the noise removing were developed. Concerning the filter input set, the noise removing methods have been divided into three classes (see Figure 1) such as temporal filtering, spatial (planar) filtering, and spatiotemporal filtering. The class of temporal filters (see Figure 1a) is referred
to temporal correlation of frames. One-dimensional filters remove noise without impairing the spatial resolution in stationary areas. In the case of large motion, the performance of temporal filters is insufficient [1, 2, 3] and the temporal filtering must be connected with motion compensation [4] so as to filter objects along their motion trajectory. However, this way is very computational
Data Loading...
