Utilizing Soft Information in Image Decoding Hannes Persson and Anna Brunstrom Dept. of Computer Science Karlstad University 651 88 Karlstad, Sweden {Hannes.Persson, Anna.Brunstrom}@kau.se Tony Ottosson Dept. of Signals and Systems Chalmers University of Technology 412 96 Göteborg, Sweden Tony.Ottosson@s2.chalmers.se Abstract—This paper explores the utilization of soft information in image decoding. Error free transmission is not always possible in today's nor future's communication networks without retransmissions. Retransmissions, however, introduce delay. To delay sensitive applications the concept of soft information, i.e. lower layer knowledge about the channel conditions, could reduce the number of retransmissions. It is not sufficient to treat the soft information and the image decoder separately. By combining soft information and the knowledge of the image structure in the decoding process image quality can be improved. A modified JPEG2000 image decoder that utilizes soft information has been developed. Experimental results with images transmitted over a simulated wireless channel show that iterative decoding with soft information can give high gains in image quality. Keywords: soft information; image coding; JPEG2000 I. INTRODUCTION In future wireless information exchange, multimedia is predicted to be one of the main applications. The multimedia users have high demands on the delivered data e.g. constraints on delay and quality. This paper addresses the problem of image transmissions over erroneous wireless channels. When transmission errors occur in a noisy channel a common technique at link and transport layers is to retransmit the erroneous or lost packets. This is, however, not always the best solution in wireless multimedia communications due to the similar erroneous behavior on the channel on the next transmission [3]. Retransmissions will also result in an inefficient use of the channel. Often a tradeoff between the transmission delay versus the quality of the received multimedia must be performed. This tradeoff causes the sender and receiver to have strategies for dealing with transmission errors. Methods, apart from retransmissions, for dealing with this type of errors in multimedia include applying forward error correction (FEC) or layered coding (see review in [12]). Many multimedia decoders that exist today support features for resilience against bit-errors (e.g. JPEG2000 and MPEG-4 [8]). The multimedia applications possess great knowledge about the structure of the information. Adding lower level knowledge about channel conditions at the application level could strengthen the error handling capabilities of the application even further. The lower level knowledge that is added in the application could be in the form of soft information. The basic idea behind the use of soft information is to make the endpoints aware of the channel conditions [2]. Involved endpoints can then adjust accordingly. Soft information could be used both at the sender and at the receiver and on different layers for different purposes. A source encoder could e.g. adjust its compression and prepare the code-stream for different channel configurations. This paper focuses on how soft information from the physical layer could be used at the application layer. Here we only consider using soft information at the receiving side. The soft information is calculated with maximum a-posteriori (MAP) estimates. The estimates are reliability measures for the received bits and are generated in the physical layer [2]. Together with the decided bits the soft information is delivered to the application. The transparency of the protocol structure is preserved with the soft information since the MAP estimates are transparent of the physical layer. As an example application the international standard for still image compression JPEG2000 (see [4]) is used. The target environment concerns unreliable and foremost wireless channels where multiple bit-errors can occur. Images that are sent over these wireless channels should contain some error resilient information to be able to detect the bit-errors. These error resilient tools are provided by the JPEG2000 standard and they are encoded at the sender side. A modified JPEG2000 decoder that combines soft information and the error resilient tools has been implemented. Experiments performed with the modified decoder and a simulated channel, where a varying number of bit-errors can occur, illustrate that image quality can be improved without making retransmissions or applying FEC. Under favorable conditions high gains in image quality have been observed. The modified decoder assumes that the auxiliary image data, i.e. the image header, is transmitted in a reliable manner. In the target environment, where the channel is prone to errors, it is a requirement that the underlying communication layers are able to overlook bit-errors in user data. In this paper we only consider the application layer and thus assume that the underlying layers (link to transport layers) are able to accept bit-errors in the payload. Transport layer modifications that accommodate payload data with bit-errors are being explored, e.g. [1] and [6]. Since the modification is limited to the receiver we can also assume control of the whole protocol stack in the mobile terminal. Further, inter-layer communication must be supported to propagate the soft information from the physical to the