Applied Mechanics and Materials Vols. 333-335

Abstract: In this paper, a moving object detection method based on image blocking and improved Particle Filter is proposed. Firstly, each frame in a video is partitioned into blocks and a low-dimensional descriptor is extracted for each block; Secondly, each block is sequentially passed through two classifiers, improved Particle Filter and temporal correlation check. Finally, foreground object is detected by integrating the classification results of all blocks. We use two-stage classifier to effectively minimize the number of false positive in the detected foreground object. Experiments show that the method proposed in this paper is feasible both in qualitative and quantitative analysis so it can be used in moving object detection effectively.

Abstract: Sub-pixel positioning is one of the key technologies which can achieve positioning better than entire pixel accuracy. The paper puts forward an improved Sobel-Zernike moment sub-pixel method based on ordinary Zernike moment method, which increase the algorithm speed while the algorithm accuracy remain basically unchanged. Then apply it to the cable sheath material thickness measurement, The experiment results indicate that the algorithm speed can be above 85% on a certain accuracy.

Abstract: Aiming at the problem that security of digital watermarking is imperfect, a new digital watermarking algorithm based on holography is put forward. In embedment watermarking stage, first, original watermarking information is divided into two parts: amplitude and phase information. Then, a hologram which will be used as new watermarking information is generated by interference of the phase information light and the reference beam, and the amplitude information will be the key when original watermarking is extracted. At last, host image is transformed by the multilevel structures wavelet transform, and the new watermarking information is embedded into its suitable wavelet domain. In extraction watermarking information stage, the original watermarking information is obtained by the extracted watermarking information combined with the amplitude information. The theory and simulation show our algorithm has excellent performance in security, shear resistance and anti interference of white noise.

Abstract: Based on SVD decomposition, an digital watermarking algorithm with the method of wavelet transform is proposed. By means of singular value decomposition to realize blind extracting, the watermarking image is embedded into intermediate frequency sub bands of wavelet composition. Finally, applying normalized cross-correlation function (NC) and peak signal to noise ratio (PSNC), it has proved that this algorithm has better invisibility and robustness.

Abstract: This work presents the design and implementation of drivers fatigue detection system based on FPGA to prevent car accidents. According to the bright pupil phenomenon, which is produced by the retina when the incident lights wavelength is 850 nm, drivers eyes can be detected easily. While acquiring the real-time video of the drivers face by camera, the system accomplishes the detection of drivers eyes by using a simplified PCNN (pulse coupled neural network) and the computation of the PERCLOS (Percentage of Eye Closure) to decide whether the driver is fatigue or not. All the designing and accomplishments of the system are based on the FPGA platform Xilinx Virtex Pro Development Board. During the experiments, the system has the ability of processing 25 frames/sec, which is the speed of collection of the used camera. Also, the fatigue detection system has good stability and accuracy.

Abstract: The segmentation and labeling of sub-cortical structures of interest are important tasks for the assessment of morphometric features in quantitative magnetic resonance (MR) image analysis. Recently, multi-atlas segmentation methods with statistical fusion strategy have demonstrated high accuracy in hippocampus segmentation. While, most of the segmentations rarely consider spatially variant model and reserve all segmentations. In this study, we propose a novel local patch-based and ranking strategy for voxelwise atlas selection to extend the original Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm. The local ranking strategy is based on the metric of normalized cross correlation (NCC). Unlike its predecessors, this method estimates the fusion of each voxel patch-by-patch and makes use of gray image features as a prior. Validation results on 33 pairs of hippocampus MR images show good performance on the segmentation of hippocampus.

Abstract: In this paper we proposed an improved watershed algorithm for the quasi-circle overlapping images of the bars end face. According to the classical watershed algorithm, which often causes over-segmentation, the improved algorithm does a series of pretreatment with the original image, such as sobel filter. With the gradient operator and mathematical morphology method, we firstly obtain the smooth image of the forced local maximum marks. Then, on the basis of the quasi-circle characteristic of the target image, we proceed to maximize the erosion with circular structure in order to prevent under-segmentation. Finally, we use the watershed algorithm to segment the gray image based on distance transform. So we can separate the target from each other to achieve the accurate counting purpose. By using the proposed algorithm in the article, we obtain satisfactory segmentation results of the quasi-circle overlapping image of the bars end face image.

Abstract: Image retrieval is generally divided into two categories: one is text-based Image Retrieval; another is content-based Image Retrieval. Early image retrieval technology is mainly based on the text, after 90 years, the content-based image retrieval emerged. So far, we mainly use image retrieval technology that based on color, texture, layout analysis and retrieval. That is: content-based Image Retrieval (CBIR). This paper review the two kinds of image retrieval methods, and introduces a variety of techniques in content-based image retrieval, we also prospect of fusion research of text and content.

Abstract: In this paper, we will propose a novel classification method of high-resolution SAR using local autocorrelation and Support Vector Machines (SVM) classifier. The commonly applied spatial autocorrelation indexes, called Moran's Index; Geary's Index, Getis's Index, will be used to depict the feature of the land-cover. Then, the SVM based on these indexes will be applied as the high-resolution SAR classifier. A Cosmo-SkyMed scene in ChengDu city, China is used for our experiment. It is shown that the method proposed can lead to good classification accuracy.

Abstract: There are a set of basic rules for stroke order in the traditional handwriting of Chinese characters, which may be listed as from left to right, from top to bottom, first inside then outside, and so on. Humanoid robot NAO is very famous in the world now, which has many sensors such as two cameras, infrared, IMU, bumper sensors and so forth. In this paper we explore how the humanoid robot NAO can write a Chinese character according to the traditional stroke order rules. As a matter of fact, the stroke order relation between any two strokes is very important in the whole stroke orders in order to lead a well writing. Therefore, the thinning algorithm is employed to propose an effective and useful method to determine the stroke order relation between any two strokes for Chinese character writing by robot NAO based on corner detection and depth-searching. Then the implementing algorithm is given for the determination of stroke order relation, and a series of experiments are carried out to validate its feasibility and effectiveness, the accuracy can achieve 90% so as to be a very satisfied result in the current stage.