The Research Of Fuzzy Support Vector Machine And Its Application In Processing Of Scene Images

Scene perception and understanding are indispensable abilities for mobile robot to navigate and explore complex environment automatically, which is dependent on the results of scene images processing algorithms. However, the nature of scene images is always random, diversified and complex. Furthermore, the prior vision information of scene images is quite often poor and the technology of object recognition in scene images is still immature. It is interesting to study how we can design efficient processing algorithms of scene images for humanoid robot.Support Vector Machine (SVM) build on the strict theoretical basis, which has been proved to prossess the remarkable characteristics of good generalization performance, the small sample, nonlinear, and high dimension. Recently, it has been successfully used to the system recognition, robot control, remote sensing image processing and economy management fields, etc. However, the problem of integrating SVM and experiential knowledge from experts is unfortunately ignored in the previous works. In many engineering applications, it is possible to obtain prior information and qualitative analysis from unseen data. Thus, it is interesting to study fuzzy theory with SVM methods, which combine the experiential logic of experts into SVM model.Some problems, which related to support vector machine, have been studied in the paper, and the proposed algorithms have been applied to process scene image. The main contributions of this paper are as follows:1) A novel Three-Domain Fuzzy Support Vector Regression (3DFSVR) is proposed, which will enhance the potentials of Two-Domain Support Vector Regression (2DSVR) to handle uncertainties. When compared with traditional two-domain SVR, the major advantage of3DFSVR is able to use the prior knowledge via the novel fuzzy domain to analyze uncertain data and signals. The Three-Domain Fuzzy Kernel Function (3DFKF) provides a solution to process uncertainties and input-output data information simultaneously, which also integrate the kernel and fuzzy membership function into a three-domain function. Definition and solution of Fuzzy Convex Optimization problem (FCOP) are presented to construct the whole theoretical framework. Experiments and simulation results show the effectiveness of3DFSVR for the uncertain image denoising.2) It is interesting to study how we can design denoising algorithm which not only can deal with the uncertainty of sample density but also take account of local similarity in images. A new Fuzzy Density Weight based Least Squares Support Vector Regression (FDW-LSSVR) denoising algorithm for humanoid robot which assigns fuzzy membership values for feature vector in order to reduce the effect of uncertainty of sample density on LSSVR is proposed. It also present a new method for the design of fuzzy membership, which is designed via fuzzy theory based on the sample density. Extensive experimental results demonstrate that our method can obtain better performances in terms of both subjective and objective evaluations than those state-of-the-art denoising techniques.3) By integrating the kernel design with Type-2fuzzy sets, a systematic design methodology of Interval Type-2Fuzzy Kernel based Support Vector Machine (IT2FK-SVM) classification for scene images is presented to improve robustness and selectivity in the humanoid robot vision. Firstly, scene images are represented as high dimensional vector extracted from intensity, edge and orientation feature maps by biological-vision feature extraction method. Furthermore, a novel Probabistic Fuzzy Kernel based Principal Component Analysis (PFK-PCA) method is proposed to select the prominent variables from the high-dimensional scene image representation. Finally, an IT2FK-SVM classifier is developed for the comprehensive learning of scene images in complex environment. IT2FK-SVM is able to deal with uncertainties when scene images are corrupted by various noises and captured by different view angles. The proposed IT2FK-SVM method yields over92%classification rates for all cases. Moreover, it even achieves98%classification rate on the newly built dataset with common light case.4) A framework of Grouping-Feature and Nesting-Kernel Support Vector Machine (GFNK-SVM) methodology is presented to achieve a more reliable and robust segmentation performance for humanoid robot. Firstly, the pixel wise intensity, gradient and C1SMF features are extracted via the local homogeneity model and Gabor filter, which would be used as inputs of GFNK-SVM model. Then, a new clustering method, which we called as Cluster Validity-Interval Type-2Fuzzy C-Means (CV-IT2FCM) clustering algorithm, is proposed to achieve sample selection by integrating a Type-2fuzzy criterion in the clustering optimization process to improve the robustness and reliability of clustering results by the iterative optimization. Finally, by integrating SVM with a novel nesting-kernel, a systematic GFNK-SVM framework is presented and its model is trained as classifier for scene images segmentation. Experiments on the BSDS dataset and real natural scene images demonstrate the superior performance of our proposed method.