As an important type of image classification task, fine-grained image classification is of great practical interest. The major challenges of fine-grained image classification come from (1) The available training datasets are lack of object localization annotations which are highly time-consuming (2) The object discriminating a class generally only takes a very small portion of a whole image. In this paper, a spatial saliency extraction (SSE) approach is proposed to crop the possible attention areas inside a whole image to improve the fine-grained image classification performance. The proposed algorithm has been applied to the recently released Chest-X-ray14 dataset [1], which is highly imbalanced and weakly annotated. The SSE method offers the state-of-the-art classification performance.

In this correspondence, we propose an approach to construct irregular low-density parity-check (LDPC) codes based on quasi-cyclic extension. When decoded iteratively, the constructed irregular LDPC codes exhibit a relatively low error floor in the high signal-to-noise ratio (SNR) region and are subject to relatively few undetected errors. The LDPC codes constructed based on the proposed scheme remain efficiently encodable

Replica shuffled versions of iterative decoders for low-density parity-check (LDPC) codes and turbo codes are presented. The proposed schemes can converge faster than standard and plain shuffled approaches. Two methods, density evolution and extrinsic information transfer (EXIT) charts, are used to analyze the performance of the proposed algorithms. Both theoretical analysis and simulations show that the new schedules offer good tradeoffs with respect to performance, complexity, latency, and connectivity

In this paper, the mean field (MF) and mixed mean field (MMF) algorithms for decoding low-density parity-check (LDPC) codes are considered. The MF principle is well established in statistical physics and artificial intelligence. Instead of using a single completely factorized approximated distribution as in the MF approach, the mixed MF algorithm forms a weighted average of several MF distributions as an approximation of the true posterior probability distribution. The MF decoding algorithm for linear block codes is derived and shown to be an approximation of the a posteriori probability (APP) decoding algorithm. The MF approach is then developed in the context of iterative decoding and presented as an approximation of the popular belief propagation decoding method. These results are extended to iterative decoding with the MMF algorithm. Simulation results show that the MF and MMF decoding algorithms yield a good performance-complexity tradeoff, especially when employed for decoding LDPC codes based on finite geometries.

Reduced latency versions of iterative decoders of low-density parity-check codes are analyzed in this paper. The proposed schemes converge faster than standard approaches. Two methods, density evolution and EXIT charts, are used to analyze the performance of the proposed algorithms. Both theoretical analysis and simulations show that the new schedules offer good performance versus complexity and latency trade-offs.

A two-dimensional post normalization scheme is proposed to improve the performance of conventional min-sum (MS) and normalized MS decoding of irregular low density parity check codes. An iterative procedure based on parallel differential optimization algorithm is presented to obtain the optimal two-dimensional normalization factors. Both density evolution analysis and specific code simulation show that the proposed method provides a comparable performance as belief propagation decoding while requiring less complexity. Interestingly, the new method exhibits a lower error floor than that of belief propagation decoding in the high SNR region. With respect to standard MS and one-dimensional normalized MS decodings, the two-dimensional normalized MS offers a considerably better performance.

Replica shuffled versions of iterative decoders of turbo codes, low-density parity-check codes and turbo product codes are presented. The proposed schemes converge faster than standard and previously proposed "shuffled" approaches. Simulations show that the new schedules offer good performance versus complexity/latency trade-offs.

Shuffled versions of iterative decoding of low-density parity-check codes and turbo codes are presented. The proposed schemes have about the same computational complexity as the standard versions, and converge faster. Simulations show that the new schedules offer better performance/complexity tradeoffs, especially when the maximum number of iterations has to remain small.