Applied Mechanics and Materials Vols. 568-570

Abstract: Pedestrians are quite visually different at multi-scale changes and view-point variations, which are crucial factors for detection. Firstly calculations of multiscale features of HOG(Histogram of Oriented Gradient) from video images are usually based on finer scale pyramid strategy, which is to figure out low-level features respectively at each scale level. This scheme has redundent information and worse real-time performance, which become urgent bottleneck in applications. A new fast scale pyramid strategy based on feature forecast algorithm was adopted according to [1], which can speed up low-level feature calculations and solve real-time problem fundamentally. Secondly a new tracking pedestrian algorithm was proposed, which combined local LBP(Local Binary Pattern) and HOG as a model description of the pedestrian target, the tracking system equation was improved according to the non-uniform motion of the pedestrian in order to enhance the effectiveness and the guidance of the particle propagation. The final experimental results show that this method is more robust than those based on single feature.

Abstract: In recent twenty years, image denoising method which is based on partial differential equations (PDE) has been developed rapidly. Due to its scientific and strong theoretical foundation, it owns accurate and stable results which can be got by efficient algorithms. But it still leaves some problems which need to be solved. The staircase effect is one of the most basic problems in the classical TV (Total Variation) model. This problem can be effectively solved by high-order model proposed in this paper. A fast and efficient numerical algorithm is designed to solve minimization problems related to the high-order model and its applications to variational image denoising are shown. The performance of the proposed model is compared with TV model and other high-order models. The algorithm used for the proposed model is also compared with Split Bregman algorithm. Some numerical experiments validate the model proposed and the algorithm designed in this paper.

Abstract: The spectral characteristic to classify the remote sensing image classification methods based on pixels of tradition, and the object oriented classification method besides the spectral information, texture feature, also includes the spatial structure of images and other information, so the classification accuracy is very high. In this paper, the remote sensing image based on object oriented classification, puts forward the classification of remote sensing image segmentation based on multiple information combination. Experiments show that, this method can overcome the pixel maximum likelihood classification based on frequent pepper phenomenon of tradition, greatly improves the classification accuracy and reliability. and has better visual effect.

Abstract: A novel face detection approach based on double skin models and AdaBoost algorithm is proposed in this paper. The image segemention of skin regions is firstly got with a fixed threshold skin model in YCbCr color space. The image segemention result is used for optimizing the parameters of Gaussian skin color model which is used for the image segmentation of the skin regions secondly. The logical operations are computed with the twice results of skin segmentation and then the coarse position of candidate face regions are got by morphological processing. Finally, the accurate face regions are acquired combined with Adaboost algorithm. The experimental results indicate that this face detection method can restrain the false number better on the premise that a proper detection rate should be kept and has better robustness.

Abstract: With the constant improvement of the living standard of the people and the increased expectations for food products of high quality and safety standards, the quality of food materials needs to satisfy the higher demand. The separation quality of the cereals and oils is one of the key factors for ensuring food quality. The traditional separating methods have many shortages such as inconsistency and uncertainty. The machine vision technology applied in the separation processing can realize the quality inspection of products and satisfy the requirements of objectivity, accuracy, consistency, efficiency and online inspection. So machine vision technology improves the separating quality of cereals and oils and realize the intelligence and the automation in the separating process.

Abstract: The wavelet analysis has some important applications in image processing, including image compression, image de-noising and so on.Wavelet analysis for two-dimensional image compression is a key aspect in the field of its applications.This paper studied the application of wavelet analysis in BMP image coding, the characteristics of wavelet coefficients and wavelet subimage, these lay the foundation for further selection of wavelet coefficients and optimize, and analyzed the theory of EZW algorithm, illustrates the better results of the applications on using wavelet theory in image processing.

Abstract: With the rapid development of Internet technologies and applications, a lot of harmful video have spread out on Internet, it is enormously harmful to stability of social and people's physical and mental health. The means to extract video caption text is studied in this paper, and improvement method of text security detection. The proposed method first classifies caption text, then compares the result of classification with the library of user’s demands to determine whether to trigger alarms, through which the aim of monitoring harmful videos could be achieved. In this method, the text detection manner calculate the polarity of sentiment words by analyzing the context of those, meanwhile considers the effect of noun, then gets the orientation of the whole texts. Experiment has shown the method can monitor harmful video effectively.

Abstract: This paper mainly for relatively small moving target, relatively open space across the network sparring sport badminton Video studied mining methods. For the basic motion feature mining method, accurate extraction of a moving object based on the mining model is based on motion trajectories using the modified geometric features to be analyzed, to obtain some of the video structural information based on statistics of the early exercise habits and issues and so on, using the improved tracking algorithm to solve the problem of missing and made a video teaching system design badminton.

Abstract: Research using image processing techniques to identify new burning point of industrial tube furnace.First, get the flame image through the design system, and then using the symmetric differencing to obtain each furnace burning point position, and finally the use of Clustering Algorithm to identify the new burning point. Through the experimental simulation show that this algorithm can avoid the complex background interference in the furnace, accurate and effective identify the new burning point.

Abstract: Stereo matching has been the focus of computer vision research. Some current researches on stereo matching algorithms were summarized, the stereo matching key techniques were analyzed; View of the current major challenges on binocular stereo matching, elaborated matching algorithm possible solutions and modification approaches; Finally, the field of technology development are prospected.