[신간의 별자리x] 우리/미술, 그리고 ‘슬픔의 박물관’

Based on the study of motion vector distribution from several commonly used test image sequences, a new diamond search (DS) algorithm for fast block-matching motion estimation (BMME) is proposed in this paper. Simulation results demonstrate that the proposed DS algorithm greatly outperforms the well-known three-step search (TSS) algorithm. Compared with the new three-step search (NTSS) algorithm, the DS algorithm achieves close performance but requires less computation by up to 22% on average. Experimental results also show that the DS algorithm is better than the four-step search (4SS) and block-based gradient descent search (BBGDS), in terms of mean-square error performance and required number of search points.

A new diamond search algorithm for fast block-matching motion estimation

Обнаружение транспортных средств на изображениях загородных шоссе на основе метода Single shot multibox Detector

The mainstream approach to image quality assessment has centered around accurately modeling the single most relevant strategy employed by the human visual system (HVS) when judging image quality (e.g., detecting visible differences, and extracting image structure/information). In this work, we suggest that a single strategy may not be sufficient; rather, we advocate that the HVS uses multiple strategies to determine image quality. For images containing near-threshold distortions, the image is most apparent, and thus the HVS attempts to look past the image and look for the distortions (a detection-based strategy). For images containing clearly visible distortions, the distortions are most apparent, and thus the HVS attempts to look past the distortion and look for the image's subject matter (an appearance-based strategy). Here, we present a quality assessment method [most apparent distortion (MAD)], which attempts to explicitly model these two separate strategies. Local luminance and contrast masking are used to estimate detection-based perceived distortion in high-quality images, whereas changes in the local statistics of spatial-frequency components are used to estimate appearance-based perceived distortion in low-quality images. We show that a combination of these two measures can perform well in predicting subjective ratings of image quality.

http://vision.okstate.edu/pubs/MAD_JEI.pdf

Most apparent distortion: full-reference image quality assessment and the role of strategy

Quantization is a process that maps a continous or discrete set of values into approximations that belong to a smaller set. Quantization is a lossy: some information about the original data is lost in the process. The key to a successful quantization is therefore the selection of an error criterion – such as entropy and signal-to-noise ratio – and the development of optimal quantizers for this criterion.

Quantization

FEANet: Foreground-edge-aware network with DenseASPOC for human parsing

In this paper, a new compensated sum of absolute difference (CSAD) for fast H.264 inter mode selection algorithm is proposed. The main idea is to determine the best inter mode based on CSAD cost instead of the rate-distortion (RD) cost. This approach can avoid most of the computationally intensive processes in the H.264 mode decision. The CSAD could solve the problem of SAD and SATD costs used in mode decision which normally bias to the smaller block size modes. It is because they are normally achieving higher prediction accuracy but consume more bit rate. Experimental results show that the proposed CSAD-based mode decision algorithm can reduce 60% to 68% of the H.264 total encoding time with negligible degradation in the RD performance.

/pdf/compensated-sum-of-absolute-difference-for-fast-h-264-inter-3skx06fun4.pdf

Compensated sum of absolute difference for fast H.264 inter mode selection

In H.264/AVC, 4 × 4 discrete cosine transform (DCT) is performed on the residual signals after intra prediction for decorrelation. Actually, residual blocks with different prediction modes exhibit different frequency characteristics. Therefore, the fixed transform matrix cannot match the energetic distribution of residual signals very well, which degrades the decorrelation performance. Fortunately, the energetic distributions of residual blocks with the same mode are relatively coincident, which makes it possible to train a universally good Karhunen---Loeve transform (KLT) matrix for each mode. In this paper, an optimal frequency matching (OFM) algorithm is proposed to train KLT matrices for residual blocks and nine KLT matrices corresponding to nine prediction modes of 4 × 4 intra blocks are trained. Experimental results show that KLT with trained matrices yields a persistent gain over H.264 using 4 × 4 DCT with an average peak signal-to-noise ratio (PSNR) enhancement of 0.22dB and a maximum enhancement of 0.33dB.

/pdf/the-training-of-karhunen-loeve-transform-matrix-and-its-4aua3pi7og.pdf

The training of Karhunen---Loève transform matrix and its application for H.264 intra coding

Content-based image retrieval (CBIR) techniques will be part of the next generation of search engines. However, even with a suitable retrieval method, real-time image retrieval from a large image database still requires extremely high computational power. In this paper, a Web-based image database indexing and retrieval system is described. It uses fast CBIR with Beowulf-Class parallel computing engine (Super Abacus) to efficiently and effectively shorten the retrieval time. Content-based image retrieval with several parallel algorithms for images which query different color spaces and the compressed-domain of the images are developed to achieve higher efficiency. Text-based query is also supported for images using a conventional annotation approach.

Web-based Beowulf-Class parallel computing on image database indexing and retrieval system

Easily trapped in local minima is one of the well-known problems in search point pattern based fast block motion estimation algorithms. This problem is especially serious in one-at-a-time search (OTS) and block-based gradient descent search (BBGDS). These two algorithms can provide very high speedup ratio but with low robustness in prediction accuracy especially for sequences with complex motions. Multi-path search (MPS) using more than one path have been proposed to improve the robustness of BBGDS, but the computational requirement is much increased. To tackle this problem, a novel multi-directional gradient descent search (MDGDS) is proposed in this paper with use of multiple OTSs in eight directions. Basically, the proposed MDGDS performs eight one-dimensional gradient descent searches on the error surface and therefore can trace to the global minimum more efficiently. Experimental results show that a significant improvement in computation reduction can be achieved as compared with well-known fast block motion estimation algorithms.

/pdf/multi-direction-search-algorithm-for-block-based-motion-2h484aj1lc.pdf

Lai-Man Po

Papers

FEANet: Foreground-edge-aware network with DenseASPOC for human parsing

Compensated sum of absolute difference for fast H.264 inter mode selection

The training of Karhunen---Loève transform matrix and its application for H.264 intra coding

Web-based Beowulf-Class parallel computing on image database indexing and retrieval system

Multi-direction search algorithm for block-based motion estimation