692 B. HRUŠOVSKÝ, J. MOCHNÁČ, S. MARCHEVSKÝ, ERROR CONCEALMENT METHOD BASED ON MOTION VECTOR … Error Concealment Method Based on Motion Vector Prediction Using Particle Filters Branislav HRUŠOVSKÝ, Ján MOCHNÁČ, Stanislav MARCHEVSKÝ Dept. of Electronics and Multimedia Communications, Faculty of Electrical Enginnering and Informatics Technical University of Košice, Letná 9, 040 01 Košice, Slovakia branislav.hrusovsky@tuke.sk, jan.mochnac@tuke.sk, stanislav.marchevsky@tuke.sk Abstract. Video transmitted over unreliable environment, such as wireless channel or in generally any network with unreliable transport protocol, is facing the losses of video packets due to network congestion and different kind of noises. The problem is becoming more important using highly effective video codecs. Visual quality degradation could propagate into subsequent frames due to redundancy elimination in order to obtain high compression ratio. Since the video stream transmission in real time is limited by transmission channel delay, it is not possible to re- transmit all faulty or lost packets. It is therefore inevitable to conceal these defects. To reduce the undesirable effects of information losses, the lost data is usually estimated from the received data, which is generally known as error concealment problem. This paper discusses packet loss modeling in order to simulate losses during video trans- mission, packet losses analysis and their impacts on the motion vectors losses. Keywords Packet, loss, error concealment, particle filter. 1. Introduction Packet data transmitted over wireless environment, e.g. WiFi, or in generally any network with unreliable transport protocol, is facing the losses of packets due to network congestion and noises of different kinds. If video signals coded with some of advanced video coding stan- dard are transmitted, these losses have severe impact on resulting video quality due to highly effective redundancy elimination in video coding process. Visual quality degra- dation could propagate to the subsequent frames due to redundancy elimination in order to gain high compression ratio. Therefore it is necessary to know in which way the packets are lost and one of the possible ways to learn about losses is creation of networks model. The impact of packet loss can be studied from re- corded measurement traces of traffic and loss patterns. To generate error process with similar characteristics as ob- served in measurements, stochastic model can be modeled [1]. The most popular examples of such models are dis- crete-time Markov chain models. The use of discrete-time Markov chain models, particularly the 2-state Markov chain model (sometimes called the Gilbert model) has been proposed in [2]. Discrete-time Markov chain models of increasing levels of complexity, including the 2-state Markov chain model have been described in [2], [3]. Obviously, the usage of Gilbert model is quite simple, but its major drawback is inability to correctly model heavily tailed error runs. In such cases, hidden Markov models with up to five states are used to model the distri- bution of error and error-free burst lengths [4]. Consequently, appropriate error control, recovery or error concealment methods, which have been developed over the times, can be chosen. On the one hand, traditional error control and recov- ery methods for data communication are focused on loss- less reconstruction of damaged video signal, but they also increase amount of data to be transmitted. However, these techniques introduce some redundancy. On the other hand, signal reconstruction and error concealment have been proposed to obtain close approximation of the original signal or attempt to make the output signal at the decoder less objectionable to human eyes [5]. Error concealment methods can be classified into three categories: 1) spatial, 2) temporal, 3) hybrid. Spatial error concealment techniques use the information from the surrounding correctly received or already concealed blocks to reconstruct damaged area. Typical representative of this class is weighted pixel averaging algorithm. Temporal error concealment techniques use the information of the corresponding blocks from the previous/successive blocks to conceal corrupted block. Typical representative of tem- poral error concealment methods is boundary matching algorithm and also methods based on Bayesian filtering theory. Hybrid error concealment techniques use the in- formation from the spatial domain as well as information from the temporal domain [5]. This paper is focused on packet loss analysis resulting in creation of packet loss model for fixed network topol- ogy, as well as for wireless topology. Consequently, cor- rupted video sequences were concealed with particle filter based error concealment.