Aysegul Cuhadar

Combined source and channel coding with JPEG2000 and rate-compatible low-density Parity-check codes

Rate-compatible low-density parity-check (RC-LDPC) codes are used to provide unequal error protection for the robust and efficient transmission of JPEG2000 compressed images over noisy channels. The total bit budget is partitioned between the source and the channel coding by using a Viterbi algorithm (VA) applied to a search trellis, and appropriate channel code rates are assigned to the source blocks. The performance of the proposed scheme is evaluated on binary symmetric channels (BSCs). Experimental results indicate that the proposed scheme compares favorably with other combined source/channel coding schemes over a variety of channel conditions and transmission bit rates. In particular, the proposed scheme outperforms similar schemes based on turbo codes and irregular repeat-accumulate codes by up to about 1.1 and 1 dB in the expected peak signal-to-noise ratio (PSNR) of reconstructed images, respectively.

Progressive Transmission of Images Over Fading Channels Using Rate-Compatible LDPC Codes

In this paper, we propose a combined source/channel coding scheme for transmission of images over fading channels. The proposed scheme employs rate-compatible low-density parity-check codes along with embedded image coders such as JPEG2000 and set partitioning in hierarchical trees (SPIHT). The assignment of channel coding rates to source packets is performed by a fast trellis-based algorithm. We examine the performance of the proposed scheme over correlated and uncorrelated Rayleigh flat-fading channels with and without side information. Simulation results for the expected peak signal-to-noise ratio of reconstructed images, which are within 1 dB of the capacity upper bound over a wide range of channel signal-to-noise ratios, show considerable improvement compared to existing results under similar conditions. We also study the sensitivity of the proposed scheme in the presence of channel estimation error at the transmitter and demonstrate that under most conditions our scheme is more robust compared to existing schemes

A fast trellis-based rate-allocation algorithm for robust transmission of progressively coded images over noisy channels

We propose a fast trellis-based rate-allocation algorithm for robust transmission of progressively coded images over noisy channels. The algorithm, which is an improved version of a similar algorithm by Banister et al., is based on the application of the Viterbi algorithm to a search trellis. This trellis is a substantially trimmed version of the one used by Banister et al.. The proposed algorithm is applied to images encoded by the set partitioning in hierarchical trees and the Joint Photographers Expert Group 2000 for transmission over binary symmetric channels. For different total bit budgets and channel parameters, speed-up factors of up to about three orders of magnitude are achieved.

Robot tracking using vision and laser sensors

We present a real-time person tracking system for a mobile robot using the information from a PTZ video camera and a laser range finder. For robust tracking, the target person model is constructed. The model includes the color, edge and size information of the tracked person. People moving in the field of view are detected by the robot via the laser range finder and verified against the target person model using the images captured by the onboard camera. A particle filter is used to estimate the position and the velocity of the moving target person for controlling the robot to follow the target. Key characteristics of the proposed system are real-time operation and ability to adapt to different moving targets. Experimental results performed in indoor environments show the effectiveness and robustness of the developed approach.

Parallel Lossless Data Compression Based on the Burrows-Wheeler Transform

In this paper, we present parallel algorithms for lossless data compression based on the Burrows-Wheeler transform (BWT) block-sorting technique. We investigate the performance of using data parallelism and task parallelism for both multi-threaded and message-passing programming. The output produced by the parallel algorithms is fully compatible with their sequential counterparts. To balance the workload among processors we develop a task scheduling strategy. An extensive set of experiments is performed with a shared memory NUMA system using up to 120 processors and on a distributed memory cluster using up to 100 processors. Our experimental results show that significant speedup can be achieved with both data parallel and task parallel methodologies. These algorithms will greatly reduce the amount of time it takes to compress large amounts of data while the compressed data remains in a form that users without access to multiple processor systems can still use.

Quadtree-based multiregion multiquality image coding

Region-based video coding schemes employed in MPEG-4 are also promising for still image coding applications where images contain a number of objects that can be encoded at different bit rates, such as compression of medical images for archiving and transmission. Motivated by this fact, in this letter we investigate multiregion multiquality (MRMQ) coding with quadtree-based wavelet coders. We present a novel scheme which addresses the region size sensitivity problem in region-based coding. The proposed method outperforms the region-of-interest (ROI) coding unit of JPEG-2000; it is possible to save 0.3 bits per pixel to attain the same ROI/background fidelity without sacrificing the additional features provided by JPEG-2000, such as resolution scalability and error resilience.

ROI coding with integer wavelet transforms and unbalanced spatial orientation trees

In this paper, we present region-based coding of medical images using integer wavelet transforms and the modified SPIHT algorithm. Our method differs from previously reported region-of- interest (ROI) coders in such a way that the region-based integer wavelet transform is used to obtain the representation of the partitioned image plane rather than differentiating the coefficients associated with each region after using the conventional wavelet decomposition. In fact, this region-based representation retains the properties of the conventional wavelet transform, and thereby facilitates the use of conventional wavelet coefficient plane coders for region-based coding. We propose a novel region-based coder based on the SPIHT algorithm. Previous region-based SPIHT coders employ conventional one-to-four parent-child binding, which accumulates coefficients from different regions within a spatial orientation tree. Alternatively, we present the unbalanced spatial orientation tree structure, which prevents the aforementioned heterogeneity in the tree, and size of which adapts to the size of the region being encoded. In addition to its superior rate-distortion (R-D) performance, the proposed coder offers region-size insensitive coding of the partitioned wavelet coefficient plane.

Multiple, arbitrary shape ROI coding with zerotree based wavelet coders

Region-base video coding schemes employed in MPEG-4 is also promising for still image coding applications where images contain a number of objects that can be encoded at different bit rates, such as compression of medical images for archiving and transmission. Motivated by this fact, in this paper, we investigate multi-region and multi-quality (MRMQ) coding based on zerotree wavelet coders. We present a novel scheme, which addresses the region size sensitivity problem in region-based coding. The proposed method outperforms the region of interest (ROI) coding unit of JPEG-2000, i.e., it is possible to save 0.3 bits per pixel to attain the same ROI/background rate-distortion performance with the proposed MRMQ coding scheme.

Error-resilient and error concealment 3-D SPIHT video coding with added redundancy

In this paper, we propose a multiple description (MD) video coding algorithm based on error-resilient and error concealment set partitioning in hierarchical trees (ERC-SPIHT). Compared to ERC-SPIHT, the novelty of the proposed approach is the injection of additional redundancy into the substreams so that the coefficients in the spatial root subband are protected highly from transmission errors. Experimental results on different video sequences show that the proposed method maintains error-resilience with high coding efficiency. In particular, our results demonstrate that the proposed algorithm achieves a significant improvement on video quality by up to 2.24 dB in the presence of a substream loss compared to ERC-SPIHT.

Parallel lossless data compression using the Burrows-Wheeler Transform

In this paper, we present parallel algorithms for lossless data compression based on the Burrows-Wheeler Transform (BWT) block-sorting technique. We investigate the performance of using data parallelism and task parallelism for both multi-threaded and message-passing programming. The output produced by the parallel algorithms is fully compatible with their sequential counterparts. To balance the workload among processors we develop a task scheduling strategy. An extensive set of experiments is performed with a shared memory NUMA system using up to 120 processors and on a distributed memory cluster using up to 100 processors. Our experimental results show that significant speedup can be achieved with both data parallel and task parallel methodologies. These algorithms will greatly reduce the amount of time it takes to compress large amounts of data while the compressed data remains in a form that users without access to multiple processor systems can still use.