文章基本信息

标题：A Novel Approach for Classification of Indoor Scenes
本地全文：下载
作者：Gagandeep Kaur ; Dr. Amandeep Verma
期刊名称：International Journal of Computer Trends and Technology
电子版ISSN：2231-2803
出版年度：2015
卷号：23
期号：3
页码：108-112
DOI：10.14445/22312803/IJCTT-V23P123
出版社：Seventh Sense Research Group
摘要：Scene classification is recently growing area of research in computer vision. A variety of approaches has been proposed for scene classification. The literature addresses the issues involved in indoor scene classification. The segmentation based approaches suffer from poor performance of segmentation and objectbased approaches involve series of complex tasks like segmentation, training a large number of classifier and recognition. In this study, a novel approach for classification of indoor scenes into multiple classes has been proposed. The proposed feature representation is entirely based on extracting structural properties of the scene images. The proposed method uses Gaussian filter in preprocessing phase to reduce noise from image followed by using morphological operations to extract edge features from image. The onevsall Support Vector Machines (SVM) learning model is employed for learning and classification. To test the performance of classification system, a database of five indoor classes i.e. bedroom, living room, dining room, office and kitchen has been taken from MITindoor dataset. The images have been taken under different under different illumination conditions and different viewpoints. The accuracy of 84% and sensitivity of 56% has been obtained for five indoor classes.
关键词：Indoor Scene Classification; Structural properties; Morphological Gradient.