Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-11-29 Cooperative journals: 《计算机应用研究》
Abstract： It is an important scientific problem to analyze visual information represented by action potentials (Spike) of brain neurons. Reconstruction of visual stimuli by Spike signals is an important approach to solve this problem. This paper provides a method to reconstruct visual stimuli from Spike signals of neurons. This experiment adopted five kinds of character images under four color backgrounds to stimulate the pigeons, and then extracted the firing rate characteristics of Spike signals from the optic tectum (OT) neurons. And this experiment constructed a linear inverse filter and a random forest reconstruction model and optimized the model parameters to reconstruct characters images under different color backgrounds. And this paper analyzed the effect of background color on reconstruction results. The results show that under optimal parameter, the average reconstruction accuracy of Character images in white, red, green and blue backgrounds using the linear inverse filter model is 0.9225�.0268, 0.9027�.0204, 0.9358�.0235, and the average reconstruction accuracy of character images using the random forest model is 0.9499�.0255, 0.9228�.0303, 0.9472�.0239, and 0.7913�.0255. The analysis of variance shows that there is no significant difference in the image reconstruction results of white, red, and green backgrounds, but the image reconstruction results of the three colors backgrounds are significantly different from the reconstruction results of the blue background image.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-10-11 Cooperative journals: 《计算机应用研究》
Abstract： With the rapid development and popularization of the Internet and multimedia camera technology, the number of various image resources has expanded dramatically. How to quickly and effectively find the user's favorite image in many image information resources has become an important issue that needs to be solved in the field of image recommendation. Aiming at this problem, this paper proposed a user-appreciated aesthetic image recommendation method, which used the deep convolutional neural network to extract the deep features of the image, and obtained an image sorting score after SVMRank, while using hand-marked image aesthetic factors (such as : hue method, image combination rule, definition and simplicity) Calculate and obtain the aesthetic characteristics of the image, get an aesthetic score, and finally perform weighted cross-validation to obtain a recommendation result that is satisfactory to the user. Experiments show that the algorithm is an effective recommendation method for aesthetic preferences.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-20 Cooperative journals: 《计算机应用研究》
Abstract： Personalized aesthetic ranking plays an import role in Image Quality Assessment and Image retrieval. In order to solve drawbacks of the missing preference from the user and low accuracy in existing methods, this paper proposed a SVMRank based model to automatically rank the input images. The framework took as input a series of specific photos that users prefer, then deployed DCNN to extract the deep features and compare them with ones that are extracted from dataset to establish an User-specific aesthetic training dataset. Later, SVMRank was used to learn a customized hyper plane to produce a user-specific personalized aesthetic ranking. The paper conducted two experiments at mean time: 1) Several users were invited to evaluate the performance of user-specific aesthetic ranking. 2) The accuracy on binary-classification was tested. The experiment results show that the framework performs well on predicting user’s preference as well as on classifying the images from high to low quality. In conclusion, the proposed algorithm is an effective personalized ranking algorithm.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-20 Cooperative journals: 《计算机应用研究》
Abstract： Nearly 80% of stroke survivors have limb disability, in which the upper limb problems are more severe. Mirror rehabilitation robot is a product of the intersection of rehabilitation medicine, robotics, biomedicine, artificial intelligence and other disciplines. It is a cutting-edge technology to solve the problem of rehabilitation training for patients. Based on the mirror mechanism, mirror rehabilitation robot is developed. This paper aims at researching and analyzing its research status, key technology route and core implementation; paper expounding its main functions and technical features. Further, this paper analyzes its bottleneck problems in the research, and puts forward corresponding solutions, in order to provide the future research direction of mirror robot.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-24 Cooperative journals: 《计算机应用研究》
Abstract： In order to keep the balance between convergence and diversity of multi-objective particle swarm optimization algorithm, the paper proposed a self-organizing multi-objective particle swarm algorithm (SMPSO) , which utilized structure of population to solve multi-objective optimization problems. The distribution of the population and non-dominated solutions founded by self-organizing map network helped to construct the particle neighborhood relations , so as to improve the algorithm’s local and global search ability by selecting non-dominated solution as leader in the neighborhood. The elite learning strategy could help population jump out local optimum by doing mutation on elite particles. The experimental results show its ability to keep convergence and diversity, and show its effectiveness to solve the multi-objective optimization problems.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-19 Cooperative journals: 《计算机应用研究》
Abstract： Aiming at the problems such as incomplete solutions, low accuracy and slow convergence speed of current algorithms in solving nonlinear equations, this paper proposed a multimodal multi-objective differential evolution algorithm. Firstly, it transformed the nonlinear equations into multimodal multi-objective optimization problems, and initialized a random population and evaluated all individuals in the population. Through the non-dominated sorting and decision space crowding distance selection mechanism, it selected half of the individuals in the population to mutate. Then it used a new mutation strategy and boundary processing method to increase the diversity of solutions. Finally, it evolved the high-quality individuals through crossover and selection mechanism until it found all the optimal solutions. The experimental results on selected test function sets and engineering examples show that the algorithm can effectively search for optimal solutions and it is superior to the other four algorithms in the number of solutions and success rate.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-12 Cooperative journals: 《计算机应用研究》
Abstract： Considering the problem that large holes in the depth map captured by Kinect, an algorithm for inpainting holes in depth map under fuzzy C- means clustering guidance was proposed. Firstly, color image and depth map were simultaneously obtained as input. Then the proposed method used fuzzy C-means clustering algorithm for color image, image clustering results as a guiding image. For the large holes in the depth map, the proposed method used the improved fast marching algorithm to inpaint the holes from the edge to the internal layer. For the discrete void points, the algorithm inpainted them by using the improved bilateral filtering. Experiments show that the algorithm can effectively inpaint holes in depth map captured by Kinect, and the restored depth map are superior to the depth map restored by the traditional algorithms in smoothness and edge strength.
Hits
Hits
Downloads
Comment