A split and merge method for detecting circular objects has been proposed in this paper. In the split stage, curve segmentation is performed by using a dominant point detection method. The segments are then classified as linear or nonlinear segments by the linearity criterion. In the merge stage, the geometric properties of circles are applied to estimate the centers and radii for nonlinear segments. The merge procedure is used to find if there is a better fitting error. The nonlinear segments are classified into several clusters. The circles will be identified if they have small fitting errors. The proposed method has been evaluated using both of the synthetic and real images. The experimental results show that the proposed method is efficient. It can detect both of the broken and occluded circular objects effectively. In addition, it is robust for the noisy images.

Aimed at computation of integer-based image quality assessments, we propose the fixed-point arithmetic operations (FXP) in this paper. Four basic integer-based operations are defined, including, fractional addition, fractional multiplication, fractional division, and fractional square root. Furthermore, four kinds of image quality assessments are measured, including, mean, variance, structure similarity, and peak signal-to-noise ratio. The experiment results demonstrate that the proposed scheme achieves low precision loss, and it needs less computing time than floating-point arithmetic operations (FOP) do in low-level embedded system.

Grading of breast carcinoma plays an important role for identification of severity of the disease and prognostic treatment in histopathology. It depends on three indices: tubule formation, nuclear plemorphism, and mitotic count by scoring based on degrees of differentiation. However, scoring on these indices often subject to intra- and inter-observer variability due to its complexities and manual classifications by pathologists.

This study aims to accurately detect lumen and gland of duct for scoring of tubule formation and nuclear plemorphism using a series of image processing techniques so that robust grade identification can be conducted in the future. To demonstrate the performance of the proposed method, H&E stained images acquired from biopsy tissues of patients were tested. Experimental results reveal that the proposed algorithms can obtain satisfactory segmentations compared to manual segmentation by an experienced pathologist. These results can provide reliable scores in the grade of breast carcinoma.

Cracks are hazardous flaws. They may be caused by material fatigue, mechanical stresses, or corrosion. Ultrasound can penetrate many materials, hence it is widely used for detection of flaws. However, the use of ultrasound is interfered by several phenomena, such as geometric interface, coarse grain, mode conversion, ringing and measuring techniques. It is practically difficult to perform evaluation directly using the originally received ultrasonic signal as false judgment may be obtained. To effectively analyze ultrasonic images, this work first uses the methods of wavelet transform to reject signal noise. The synthetic aperture focusing technique (SAFT) which is similar to sensor fusion, is then applied to reduce the interference and detect the existence of flaws. We emphasis on crack flaws that are located on the edge of materials. The flaws were manually created in our experiments. Extensive analysis are given with respect to various crack length and depth. The results provide special observation of crack flaws.

Rain streak in an image can degrade the human vision, as well as the image's quality. However, the rain removal of a single image is a challenging problem, because the rain is moving fast and may become torrential. In this paper, a single image rain removal process based on the non-negative matrix factorization is proposed. In the proposed method, the rain image is decomposed into a low-frequency part and a high-frequency part by a Gaussian filter. Therefore, the rain component, which is usually in the middle frequency, could be discarded in high and low frequency domains. In this paper, the non-negative matrix factorization (NMF) method is applied to deal with the rain streak in the low frequency; while in the high frequency part, the concept of Canny edge detection and block copy strategy are utilized separately to remove the rain hidden in high frequency and improve the image quality. By comparing with the state-of-the-art approaches, our proposed method does not need the extra image database to train the desirable dictionary, but still reaches similar results.

A novel steganographic algorithm based on ZZW construction is proposed to improve the steganographic embedding efficiency. Low-density generator-matrix (LDGM) embedding is an efficient embedding technique for embedding a large number of messages into host. We shall propose a modified ZZW construction to improve the embedding complexity and efficiency. We use LDGM embedding techniques instead of the first channel of ZZW construction and estimate the suboptimal parameters for the bias propagation (BiP) algorithm of LDGM embedding. By using this construction, we are capable of constructing a new embedding structure with small payloads from original embedding structure with large payloads. Experimental results show that the embedding efficiencies of our construction are close to the theoretical upper bound.

Facing a large number of homogeneous-content videos shared on the internet, it is essential to seriate videos with prior information. In this paper, we propose fast video seriation (ViSA) scheme based on audio-visual features. The features, mel-frequency cepstral coefficients (MFCCs) and color histogram are extracted as audio feature and visual feature, respectively. Principal component analysis (PCA) is exploited for dimensionality reduction as well as time consumption reducing during feature matching. Subsequently, the proposed fast feature matching computes difference between features, in the meantime, estimates time difference between two videos. The chronological ordering establishes relationship among videos, and then seriates videos by synchronizing video time. The experiment results will show the proposed scheme reduces more 50% computing time than brute force approach does in feature matching. The error of time difference between groundtruth and the estimated result is very small. The experiment results will demonstrate that the proposed scheme is efficient in video seriating.

Determining the gender of a person in a given image or video is an interesting problem. This is a two-class pattern recognition task which is very useful in many potential real-life applications. As human faces provide important visual information for gender perception, a large number of studies have investigated gender classification from face perception. In this paper, we present a method which uses Center-Symmetric Local Ternary Pattern and Local Sign Directional Pattern for feature extraction to identify the gender from the facial images. The classification is performed by using a support vector machine, which had been shown to be superior to traditional pattern classifiers in gender classification problem. Experimental results on the FERET database are provided to illustrate the proposed approach is an effective method, compared to other similar methods.

This work developed a robust 2D to 3D digital video conversion technique that allows traditional 2D video content to be displayed as 3D video. The main principle of this work is to infer an accurate depth map for each frame from the 2D video sequence as with the inter-frame consistence. First, we infer three depth maps for each frame: initial block-wise depth map, initial pixel-wise depth map, and region-wise propagated depth map. Then, Kalman filter is introduced to fuse three depth maps to a final depth map. Once the depth map of the 2D video is obtained, we can generate the left and right images with it through DIBR algorithm, and display it on 3D stereoscopic display.

In the studies related to Multiple Description Coding (MDC), applying zero padding not only improves the spatial correlation between the descriptions but also increases the size. However, as we pad more zeros, the spatial correlation does not improves linearly. Therefore, the distortion is not reduced as the same amount as the size increases. In this paper we focus on how to efficiently find an acceptable trade-off between the number of padded zeros and the increase of distortion. Based on a MDC method named “Multiple Description Coding with Zero Padding based on spatial Quincunx distribution (QZP-MDC)”, in this paper we propose a distortion estimation, which is “Standard Deviation for Distortion Estimation (STD-DE)” to efficiently calculate the distortion with specific padding direction and the number of padded zeros. To search for the acceptable the number of padded zeros, we design two search methods, Naïve Search and Best Division Search, based on the proposed rate-distortion trade-off indicator. Our experiment results show that errors of distortion estimation by STD-DE are all less than 2%, and the executing times by best division search is about twice faster more than those of naïve search for 11 test sequences. To summarize our work, we confirm that the best division search on STD-DE can find the acceptable trade-off between rate and distortion in low computational complexity.

In this paper, a two-level near-duplicate video detection method based on invariant moment was proposed. To reduce the computational complexity of near-duplicate video detection, a coarse-to-fine approach was adopted in the proposed method. The proposed method is composed of key-frame selection, invariant moment calculation, feature point extraction and matching, similarity measurement, and near-duplicate classifier. After key-frame selection, the proposed method coarsely finds the corresponding frame pairs based on invariant moments. For each chosen frame pair, SURF is used to find the corresponding point pairs between the query frame and the test one. After feature-level, spatial-level, and temporal-level similarity measurement, we can decide whether the query video clip and the test one are near-duplicate. The experimental results show that the proposed method can effectively detect near-duplicate videos. In addition, the proposed method has good performance against possible operations, re-scaling, and frame-rate change.

This paper aims to investigate a multi-pose face recognition system based on an effective design of classifier using SURF feature. In training phase, the proposed method utilizes SURF features to calculate similarity between two images from different poses of the same face. Face recognition model (GMM) is trained using the robust SURF features from different poses. In testing phase, feature vectors corresponding to the test images are input to all trained models for the decision of the recognized face. Experimental results are presented to demonstrate the comparative performance of the proposed method.

In this research, we capture body movements, facial expressions, and voice data of subjects in the presentation scenario using RGBD-capable Kinect sensor. The acquired videos were accessed by a group of reviewers to indicate their preferences/aversions to the presentation style. We denote the two classes of ruling as Period of Like (POL) and Period of Dislike (POD), respectively. We then employ three types of image features, namely, animation units (AU), skeletal joints, and 3D face vertices to analyze the consistency of the evaluation result, as well as the ability to classify unseen footage based on the training data supplied by 35 evaluators. Finally, we develop a prototype program to help users to identify their strength/weakness during their presentation so that they can improve their skills accordingly.

Tooth contour is essential for orthodontic evaluation. In this study, a method for tooth segmentation from dental panoramic radiographs is proposed to assist orthodontists to obtain reliable therapy planning. The proposed method-locates the position of the target tooth, optimizes the scale, translation, and rotation of a tooth model, and deforms the model based on an energy minimization way. Visual inspection, qualitative analysis, and quantitative analysis were used to evaluate the results. The experimental results show that the accuracy of the proposed method is similar to manual delineationby experts.

Due to the rapid development of social networks, users used to share information, such as blogs with friends in public throughthe social networks. Blogs, one of the main platforms for social networks, are still like isolated island. To overcome this problem, interlinks are used to link blogs. The paper extends the platform we developed, called Blog Connect, to automatically create interlinks among blog posts. The queried keywords collected by Blog Connect are used to search the corresponding category from DBpedia. In the experiments, the precision is above 50% to show its feasibility.

Automatic inspection techniques have been widely employed to achieve high productivity while ensuring high-quality products in steel making industry. In this paper, a vision-based detection framework for automatically detecting different types of steel billet surface defects is proposed. The defects considered in this study includes cratches, corner cracks, sponge cracks, slivers, and roll marks. In the proposed framework, to improve the quality of image acquisition for billet surface, two preprocessing techniques, i.e., automatic identification of ROI (region of interest) and HDR (high dynamic range)-based image enhancement techniques, are proposed. Then, DWT (discrete wavelet transform)-based image feature is extracted from the image to be detected and fused with the other two extracted local features based on variance and illumination to identify each defect on the billet surface. Experimental results have verified the feasibility of the proposed method.

With the advance of multimediatechnology and communications, image and video data over the Internet contribute to the sea of media. Methods on how to search and findout user desired media contents becomesimportant. In this paper, we proposed an efficient content-based video retrieval (CBVR) method that integrated color, texture and SIFT-BOW (Bag of Word) image features for robust and higher retrieval performance. Experimental results showed that our proposed CBVR method can well manipulate both global and local features for retrieval and can outperforms the previous methods by 12.5% and 19.7 %in precision performance that adopts single and multi-feature types.

Wavelet transform has been received a great interest because it provides a multi-resolution representation of images. Wavelet transforms have been shown to achieve high compression ratios while maintaining very good image quality, due to the fact that edge characteristics of images can be well preserved by this method at low bit ratesBy way of vector quantization technique, the wavelet transformed coefficients can be further compressed. For real time image compression, achieving high data compression ratio and coding efficiency at the same time are both important. In this study, we have proposed a preprocessed quadtree segmentation method for Wavelet transform image coding. Quadtree segmentation algorithm is used to divide a given image, where regions with image detail will be segmented into blocks with smaller block size, and the background of the image will be assigned larger block size. Each image sub-block is coded by Wavelet transform scheme to gain a better compression quality. After the wavelet transform, to select appropriate size for codebook, bit allocation assignment is applied associated with the variance of each subband image block. We have also adopted rate-distortion concept to adjust the compressed bit rate. For this proposed compression scheme, simulation results show that we can achieve acceptable visual quality and high compression ratio simultaneously. Furthermore, due to the small size of the codebook, we are able to reduce the computational time. System performance analysis is demonstrated in this study.

Due to the increase of large-scale videos, efficient and effective corner detectors are necessary to facilitate video processing and the underlying applications. Designing a good corner detector often faces an accuracy/efficiency trade-off problem. To address this problem, a highly efficient and effective approach named spatial-temporal FAST corner detector is proposed to detecting corners for videos. Our approach considers both the spatial and temporal evidence to extract corners, each of which is an extreme when comparing it with the pixels on the surface of the spatial-temporal tube centered on it. To efficiently obtain spatial-temporal FAST corners, a two-step procedure is adopted. The FAST detector [1] is firstly used to filter out most non-corner pixels. Then, temporal consistency is checked for FAST corners to identify if these corners are spatial-temporal FAST corners. As a result, the detected corners are spatial-temporal coherent. In the experiments, seven the state-of-the-art corner detectors are compared. The results show the superiority of our approach in both repeatability and efficiency.

High efficiency video coding (HEVC) is the state-of-the-art video coding standard and better than the previous video coding standard H.264/AVC on coding efficiency. Under the similar image quality, HEVC can reduce data amount in the bit-stream around 50% compared with H.264/AVC. In terms of intra coding, HEVC can reduce data amount around 36% on average with respect to that of H.264/AVC. The expense of these improvements is a significant increase in the computational complexity. To reduce the computational burden, we propose an efficient intra coding algorithm based on edge information and spatial correlation for high efficiency video coding. We utilize the edge information applied in MPEG-7 and analyze the correlation of intra modes on the spatial domain to reduce the search numbers of Coding Unit(CU) and Prediction Unit(PU). In addition, our algorithm uses the rate-distortion cost(RD-cost) correlation between the current CU and adjacent CUs to early terminate computation for CU partition. Experimental results show that our algorithm can save 31.66% total time on average in HEVC encoder and maintain great video quality and similar bit-rate compared to HM11.0, and perform better than previous work.

The newest standard of video coding, the high efficiency video coding (HEVC) achieves significantly better coding efficiency than all existing video coding standards. HEVC adopts the quadtree-structured coding unit (CU) to improve the coding efficiency. Each CU node in quadtree will be traversed by depth first search process to find the best coding tree unit (CTU) partition. Although this quadtree search process can obtain the best CTU partition, it requires a very high computational complexity, especially the motion estimation (ME) process in interframe coding. To alleviate the encoder computation load in interframe coding, a fast CU depth range decision (FCUDRD) method is proposed by reducing the depth search range. Based on the depth information correlation between temporal-spatial adjacent CTUs and the current CTU, some depths can be adaptively excluded from the depth search process in advance. The best depth of current CTU is determined by an intersection between temporal predicted depth ranges by 9 Gaussian weightings from encoded blocks and spatial predicted depth ranges by 4 best weightings from encoded blocks, separately. Experimental results show that the FCUDRDcan achieve time improving ratio (TIR) about 55.73% for RA and 65.77%for LD on average, respectively, as compared to the HM8.1 with insignificant loss of image quality.

High Efficiency Video Coding standard (HEVC) can achieve higher compression performance than the previous video coding standards. HEVC employs a quad-tree based block partition scheme to handle the potential challenge of compressing the ultra high definition (UHD) videos. The quad-tree based partition segments each input frame into a series of variable-sized blocks, and tries to find the best partition that encodes the frame blocks with the optimal rate-distortion performance. To determine the block size or depth level of partition, a highly computational rate-distortion optimization process is involved to compute the best partition results in terms of minimum rate-distortion cost (RDcost). This paper proposes a novel fast CU partition algorithm based on the feature of rate-distortion variation during HEVC partitioning each block to eliminate the unnecessary partition and RDcost computation. Experimental results show that the proposed algorithm can reduce 17% to 32% execution time of HEVC encoder with negligible loss of performance.

Structured lighting is an approach that has been widely utilized for 3D reconstruction in recent years due to its reliability, simplicity and effectiveness. In this paper, we propose the mounting of a structure lighting system on a gantry to improve its flexibility, efficiency and accuracy. The proposed system is able to obtain more accurate 3D models of objects with reduced time. The paper also proposes calibration methods for the system, which calculate the correspondences between the coordinate systems of the camera, the light source and the rotational axis. Parameters obtained from calibration are used to calculate the 3D data from the acquired 2D images. Experiments have been performed to demonstrate the effectiveness of the proposed method when applied to the 3D reconstruction of human face.