Applied Mechanics and Materials Vols. 333-335

Abstract: The action recognition based on local spatial-temporal feature has attracted more and more attentions. The codebook used in the recognition is usually generated by k-means and the code words are considered to have the same importance. But in fact, it isn’t true. So in this paper, we propose two strategies to measure the importance of different code words and improve the HIK-Kmeans algorithm .The experiment result shows that our improved algorithm increases the accuracy of action recognition.

Abstract: Stereo matching of the disparity discontinuous boundaries and weak texture regions is still a problem of computer vision. Local-based stereo matching algorithm with the advantages of fast speed and high accuracy is the most common method. In order to improve the matching accuracy of the mentioned regions,a stereo matching algorithm based on edge feature of segmented image is proposed. Firstly, the reference image was segmented by Mean-Shift algorithm. Then, support window was dynamically allocated based on edge feature of segmented image. Finally, the disparity distribution of support window was adjusted by introducing weighting factor. The experimental results show that this algorithm can reduce noise and effectively improve the matching accuracy.

Abstract: Classic automatic white balance algorithm always been invalidity when there are large color-blocks or less of highlights points occurred in cast images. In this literature, an improved automatic adjustment algorithm based on image segmentation is proposed to resolve the problem mentioned above. First, color images were transformed to HSV color space and low saturation area was segmented from S channel. And then, adjustment parameters were calculated by selected points. Experimental results show that the algorithm can effectively correct varied cast image with low computational complexity, and are suitable for various scenarios.

Abstract: In some business applications, the rough content of a digital image can be accessed through public channel, but the detailed image can only be achieved by some interested buyers who pay for the key. Reversible visible watermarking provides a kind of protection, but covers parts of image. In this paper, we utilize the relationship between embedding capacity and visual distortion, and proposed a novel image degradation scheme for copyright protection based on reversible difference expansion watermarking. Experiment results show that our scheme degrades the images uniformly, and the general content is still available. And the legal users with key can restore the original image with full details and extract the copyright proof.

Abstract: The image matching is key technology of image processing applied to many fields. Image matching based on invariant features was the hot spot in image matching research recently. SIFT is one of the most effective scale, rotation and illumination invariant features. However there are a lot of wrong match pairs produced by original SIFT algorithm. Two specific types of wrong matches are analyzed and corresponding eliminating methods are given. Aiming at general wrong matches eliminating, a method based on similar triangles is proposed. Experiments are carried out, and the results confirm that: compared with the art of state, the proposed method is faster; it can eliminate wrong matches cleanly and preserve correct matches as well.

Abstract: To improve the efficiency and accuracy of the conventional SIFT-TPS (Scale-invariant feature transform and Thin-Plate Spline) method in deformable registration for CT lung image, we develop a novel approach by using combining SURF(Speeded up Robust Features) and GDLOH(Gradient distance-location-orientation histogram) to detect matching feature points. First, we employ SURF as feature detection to find the stable feature points of the two CT images rapidly. Then GDLOH is taken as feature descriptor to describe each detected points characteristic, in order to supply measurement tool for matching process. In our experiment, five couples of clinical images are simulated using our algorithm above, result in an obvious improvement in run-time and registration quality, compared with the conventional methods. It is demonstrated that the proposed method may create a new window in performing a good robust and adaptively for deformable registration for CT lung tomography.

Abstract: In order to solve the problems of the cracked and adhesive characters of license plate caused by the plate frame, rivet, light intensity, a novel character segmentation method based on character contour and template matching was presented. In the proposed methodology, the license plate image contrast was enhanced by the adaptive gray stretch method. The cracked and adhesive characters were accurately extracted owing to the spatial scalability of the contour. Then the found characters were matched according to the adaptive templates. The characters were segmented once more by the best template and then residual characters were complemented and fake characters were removed. A large number of character segmentation experiments under different illumination conditions were made. The results show that the method has a strong robustness and practicability.

Abstract: This paper presented a high-speed image acquisition system based on PXI bus and Low Voltage Differential Signaling (LVDS) technology, with the characteristics of mass data, high-speed of data acquisition in the field of modem technology and scientific research. It expatiated the whole frame, and emphasized the design and implement of the system. It simulated the Ping-Pong operation of image data transmission, and tested the performance of the system. The results of test indicate that the high-speed image acquisition system has good ability of acquisition and transmission, and it can satisfy the engineering applied demand.

Abstract: Because LBF model is sensitive to the initial position and regional model may causes excessive segmentation or inner cavity. The paper adds adaptive distance keeping level set method on LBF model. Evolution curves reduce limitation to the initial position. Meanwhile, an improved area weight coefficient makes evolution curve move inward or outward adaptively according to image information and it can detect object boundaries when it is in a region with intensity homogeneity. The method enhances capability of capturing boundary concavities. In addition, xconv2 instead of conv2 function during Matlab programming for improving the evolution speed in some sense. The experimental results show that the proposed method can get local characteristics of the brain.

Abstract: Image annotation is one of the important technologies in image retrieval and semantic analysis. To overcome the estimation and efficiency problem in CMRM model, we proposed a Visual Concept Distribution based annotation model which estimates the probability through the Visual Concept Set. Experiment results shows that our approach outperforms three classical annotation models (CMRM, CRM and PLSA-WORDS) and closes to the complicated PLSA-FUSION model.