"Geometric-structure-based Directional Filtering for Error Concealment in Image/Video Transmission"

SPIE Wireless Data Transmission at Information Systems/Photonics East'95 , vol. 2601, pp. 145-156, Oct. 1995.

ABSTRACT

The problem of recovering lost data in coded images due to imperfect communication channels has attracted much attention. This problem becomes more important in reliable interactive video communication over wireless network, since the time varying capacity of wireless channels can cause severe error bursts or dropouts. Forward Error Correction and Automatic Repeat Request have been found to have some practical limitations, especially in the wireless context. As an alternative, error concealment techniques take advantage of the spatial or temporal correlation of images and videos without incurring much overhead and delay.

A good spatial interpolation is considered as fundamental for hiding the effect of missing blocks in still images and video frames. In fact, it has been argued that in wireless communication, intra-frame coding, thus spatial interpolation, may be preferred in order to avoid severe inter-frame error propagation and to reduce the hardware complexity. This paper proposes a novel spatial directional interpolation scheme which makes use of the local geometric information extracted from the surrounding blocks. Specifically, two nearest layers of the surrounding pixels of the lost block are used to extract the geometric information and to interpolate the missing pixels. Statistics of the local directional structure is modeled as a bimodal distribution. The two nearest surrounding layers are converted into binary pattern which illustrates the local geometric structure. A measure of directional consistency is employed to resolve ambiguity of possible connections of the transition points on the inner layer. The transition lines can be specified within one-pixel accuracy, unlike other directional filtering schemes which usually filter along only one single direction chosen from a finite candidate set. The new approach produces results that are superior to that of using reference [1], as illustrated in Fig. 1. The PSNR is also superior. The computation requirement is much reduced from that of references [2]. It is observed that local structures such as edges, streaks and corners are well preserved in the reconstructed image.

Keywords

Error Concealment, Directional Filtering, Packet Video, Wireless Transmission.

Procedure of the proposed scheme

Click here for a sample procedure (in gif format)

Results

LEFT: reconstructed image using the algorithm in [1]

RIGHT: reconstructed image using our scheme

Fig. 1: Isolated block lost case

(a) Original image

(b) Image with 25% block lost

(d) Reconstructed image with the proposed scheme, PSNR = 33.73 dB

Fig. 2: Contiguous block lost case

(a) Corrupted image

(b) Reconstructed image with all isolated lost blocks recovered

[1] K. Jung et al., "Error concealment technique using projection data for block-based image coding," Proc. SPIE Conf. on Visual Commu. and Image Proc., vol. 2308, pp. 1466-1476, 1994.

[2] H. Sun and W. Kwok, "Concealment of damaged block transform coded images using projections onto convex sets," IEEE Trans. Image Processing, April 1995.

Paper

· Postscript version of the paper (2.2 M)

Other Applications

· low bit rate coding

· rate shaping for video transmission

Last update: April 8, 1996.

wzeng@ee.princeton.edu