In order to extract the complicated features for online reviews sentiment analysis, an online reviews sentiment analysis model based on rough sets is presented. It analyzes the traditional word bags features, and the regular collocation features are confirmed to play important role in identifying the reviews sentiment. The regular collocation features as rough rules extracted are incorporated into the sentiment analysis model, and the results on real hotel online reviews show the performance improvement under both Support Vector Machine(SVM) model and Naive Bayes model.
This paper proposes a method of abrupt cut detection based on motion information. Quantitative angle histogram entropy and average magnitude are defined to be the metrics of frame differences from the motion vector filed calculated by block-based motion estimation. Adaptive threshold strategies are designed to identify candidate abrupt cuts. False cuts caused by flashes and gradual transitions are eliminated by using a temporal window. In the process of key frame extraction, different algorithms which have different limitations of the number of extraction are selected adaptively according to the extents of similarity of video frames content. Experimental results show that the method has high accuracy and robustness.
ISO14299 specifying unified diagnostic services and CAN based vehicle fault diagnostic protocol ISO15765 are introduced, and the relationship between ISO14229 and ISO15765 and the architecture of ISO15765 are analyzed. The overall design of the ISO15765 protocol stack is provided in the paper. The design of the ISO15765 protocol stack uses layered and interface method. Lower layers provides services to higher layers. Modular design is adopted for developing internal operation. Experimental results show that the protocol stack satisfies ISO15765 standard and general fault diagnosis requirements.
Isometric Projection(IsoP) algorithm aims at preserving the distance among samples at the most, without taking into account the sample class discriminant information. Aiming at this problem, a face recognition algorithm based on Discriminant Isometric Projection(DIsoP) is proposed. The algorithm is based on IsoP and the Maximum Scatter Difference Criterion(MSDC) is introduced to its objective function. Therefore, an optimal objective function is obtained. After being embedded into a low-dimensional subspace, the samples of same class maintain their intrinsic neighbor relations, whereas the neighboring samples of the different class are far form each other. Experimental results on ORL face database demonstrate the effectiveness of the proposed algorithm.
H.264 encoder is a general video encoder with the highest compression efficiency. However, to apply it directly into real-time screen video encoding brings the problems, such as high CPU occupancy rate and ineffectiveness in limiting the peak bitrate. To solve the problems above, in one hand, it integrates a Mirror Driver-based changed region computation step into the classic H.264 encoder, X264. After that, it can find the unchanged macroblocks and then quick-find optimized prediction modes. On the other hand, it proposes an adaptive frame rate constant quantization parameter rate control algorithm, which limits peak bitrate by reducing the local frame rate. Experimental results show that the schema reduces CPU time by 30%, strictly controls the peak bitrate and finally achieves 3 dB~8 dB PSNR improvement.
From the view of computer vision, the particular quantitative analyses and evaluations about rotational, scale and affine invariance of standard model of object recognition in cortex are made based on textured images. The original textured images are scaled, rotated and affinely transformed respectively. Biological visual features of these transformed images, also called Standard Model Features(SMFs), are extracted by the standard model. The SMFs are used to classify the images. After classification, the area under curve is utilized to quantitatively analyze and evaluate whether the standard model has invariance or not. Compared with the Local Binary Pattern(LBP) feature, a great deal of experiments show that the biological visual features extracted by the standard model have superior scale, rotational and affine invariance to textured images.
Existing research methods cannot fulfill the service selection or service composition according to user’s individual requirements. Aiming at this problem, this paper proposes a service composition algorithm based on service quality information. Through defining the matching degree of service interfaces and considering service quality information, it converts the degree of service composition into the weight of the edge in directed graph. The service composition problem converts into the shortest path problem. Analysis result of example is given to demonstrate the effectiveness of the algorithm.
A new combination method called gray correlation analysis and multiple linear regression method is proposed to establish the optimized selection model of measuring points for rainfall. A practical case is given to prove the reasonability and feasibility of this optimized selection model, and six critical points are selected from twelve points. Result shows that this model is effective, reliable and feasible, by which critical points can be picked out optimally, and the interaction between measuring points can be removed, accordingly accuracy and robustness improve greatly.
H.264 video coding standard has high computing complexity, and is difficult to meet the high-definition video in real-time encoding. In order to solve this problem, this paper proposes a parallel coding algorithm on heterogeneous multi-core platform DM6467. By analyzing the dependence between different hardware accelerators and the characteristics of different levels of memory, a macroblock level parallel scheme is designed. Further optimization is carried out by merging the pipeline of adjacent slices and balancing the tasks assigned to Digital Signal Processor (DSP) and ARM cores. Experimental result shows that the efficiency is improved by 18%, and 1080P real-time H.264 video encoder can be realized on DM6467 platform.
File carving is the important techniques about digital crime forensic investigation. In order to acquire digital evidence from the destroyed disk media, a feature extracting algorithm about document fragment is proposed which can identify various document fragments about its entropy value. Based on this algorithm, fragment file carving algorithm is designed by combining the logic property of disk cluster. Results show that compared with other file carving algorithms, the algorithms can carve out more files.