Applied Mechanics and Materials Vol. 678

Abstract: Nature Scene classification is a fundamental problem in image understanding. Human can recognize the scene instantly after only a glance. This is mainly because that our visual attention is easily attracted by the salient objects in scene. And these objects are always representative in the natural scene. It is unclear how humans achieve rapid scene categorization. But this kind of high-level cognitive behavior can be reflected by the eye movement. To identify this ability, we propose a model with the guidance of eye movement. It combines the bag of words (BOW) and spatial pyramid matching (SPM) methods to train and test our model on support vector machine (SVM). The eye movement experiments were employed to validate our model. We found that the subjects could recognize the scenes correctly even if given only a few saliency patches with less than one second. These results suggest that the eye tracking saliency patches play an important role for human scene categorization.

Abstract: Edge detection plays an important role in medical image processing; its accuracy directly affects the diagnosis and treatment of the disease. In view of the shortcomings of the traditional Sobel algorithm, an improved Sobel edge detection algorithm is proposed in this paper. Algorithm increased 135 º and 45 º direction templates, used the local gradient and standard deviation to filter and strengthen the gradient of initial gradient image. Experiments show that the image edge detected by the new algorithm is relatively accurate, complete and clear compared with the traditional Sobel algorithm, which verifies the effectiveness of the new algorithm.

Abstract: In recent years, the port occurred on many the theft cases that boats moored ships to implement embarkation at night, which seriously affected the economic benefits and the safety of the persons on the ship. In order to all-weather real-time monitor the condition around the ship, and to find the abnormal behavior of moving targets in time, the use of infrared video sensor is proposed for mooring the condition around the ship, and intelligent analysis for infrared video. First of all, through the semi-threshold method, Canny operator and connected component labeling for target detection, and to draw the target's movement track through the centroid coordinates, and then analyses the layout scheme of infrared video sensors, finally, determine whether the moving target near to the ship based on non-calibration binocular distance measurement, such as against by suspicion, alarm immediately. The experimental results show that this method is simple and effective, it can well realize to monitor the moving target in port, and to early warning at night.

Abstract: Facial region detection has broad application prospects, but existing human face region detection methods have some rigorous requirements for the light conditions. Error detection about face area caused by the poor light conditions has a great bad effect on the follow-up processing, such as face recognition, fatigue degree evaluation based on visual. So face region detection in complex lighting conditions has always been the difficult problem. Therefore a self-adaptive illumination compensation method for color images has been proposed. Select the color face images database of the California Institute of Technology to test the method dealing with the face region detection by original image and the image after illumination compensation. In the simulation experiment, the method of Illumination compensation can effectively improve the detection accuracy. Lay the foundation for driver fatigue detection based on visual.

Abstract: This paper presents a novel approach to detect unattended and removed objects from a single fixed camera video for surveillance applications. In this paper two backgrounds rather than any tracking information are employed to detect static foregrounds. Each of the backgrounds is modeled by three Gaussian mixtures. We make a subtraction between two foregrounds extracted by the background subtraction method to initially determine the static regions. Then the region-level density analysis and texture information are combined to remove false static foreground regions caused by illumination changes and objects that are in the static-moving transitions. Finally, classification of the unattended and removed is determined by a method that compares color histograms of the static foreground zone and its extended areas in current frame. Experimental results show that our method can work well in simple scenario as well as complex environments with an unsmooth background.

Abstract: Fire image detection is a kind of important means of early prevention of fire. The current fire image detection is focus on the feature extraction, the concrete algorithm, etc., there is the lack of unified scheduling and the realization of automatic mechanism in the feature extraction of smoke or flame, the integration of all stages and other aspects of cohesion, and there is no uniform model in terms of integration. In addition, the traditional fire image detection system is facing problems such as high investment and insufficient computing capacity. To effectively reduce the detecting and early warning time, in this paper, fire image detection system based on the cloud workflow is studied. In view of the smoke image detection, we have integrated the various key of discrete nodes in fire detection through cloud workflow technology, and connect effectively the other related nodes. By building cloud workflow architecture, we organize and ensure the operation and regulation of the key nodes and their child nodes of child nodes. Through the experimental test, the method reduce the time in the whole process of fire detection, also won't reduce the accuracy of the original mode of fire detection.

Abstract: This paper presents a novel image acquisition technology for colorless raised characters. First, a grating image modulated with the height information of raised character is captured through a vision system; then, the Fourier transform method is used to get the phase of the grating image; finally, a rational design phase unwrapping method is designed to reconstruct the grayscale image of raised characters. Experiments show that the proposed method can get the well-separated images of raised characters.

Abstract: This paper introduces a new correction method for laser scanner system. Based on image recognition, the laser marking result which includes a crossed matrix is grabbed as matching pattern. A algorithm is presented to locate all crosses and measure their positions , and a correction data is generated automatically for laser scanner system. This method makes the correcting process be much faster and more accurate than conventional manual measurement way.

Abstract: We propose a pedestrian detection approach based on bag-of-visual-words and SVM method. The image feature extraction and representation are extremely challenging tasks in pedestrian detection approach, which could impact the performance of pedestrian detection. In this paper, we propose that visual vocabulary is built by clustering SIFT features of image to visual words. Classification is taken using the support vector machine (SVM), for SVM having good non-linear function learning and generalization capability solid. Numerical experiments in the evaluation of INRIATREC pedestrian data sets and the action movies demonstrate that our method shows better performance.

Abstract: A method on the feature extraction of coal and rock character recognition was mainly put forward based on image gray level distribution and gray scale average value. In order to improve the recognition effects of the image, cropping, gray level transformation, contrast enhancement, median filtering and other preprocessing work were applied individually on the raw image of coal caving and rock caving acquired from mechanized top caving face, then gray histogram of image signal of coal and rock was abstracted and the gray scale mean were calculated. The results shows that (1) the range of gray scale of top coal caving image is mainly between 10-100, and the range of gray scale of top rock caving image mainly between 90-220, (2) the gray scale mean of top rock caving image is around 130, far higher than the gray scale mean of top coal caving image of 66.