Indoor scene recognition systems based features and objects

Abstract

Indoor scene recognition , also called scene classification and identification , refers to the process of labeling the elements of the newlinegiven input scene image based on the contents. Basically there are two newlineapproaches to design an Indoor Scene Recognition System (ISRS): i) Featurebased newlineand ii) Object-based. Due to its wide applications, scene recognition newlinehas gained great research attention over the past decennial. Even though newlinedifferent methods have been proposed in the literature, there is no consensus newlineon the type of classification in a more prefect manner. Also performance of newlinethe scene recognition systems is found to be less when compared with the newlineprocess involved in it. Hence in this research work two indoor scene newlinerecognition systems are designed based on features of the given scene and newlineobjects of the scene that can provide higher performance. newlineThis research work initially proposes a new novel ISRS based on newlinefeatures of the given scene image. These features are extracted from the low newlinelevel primitives, namely homogeneity, edge and texture present in the scene newlineimage. To extract these features, the proposed system utilizes the well-known newlineorthogonal polynomials model. A new block decomposition model is newlinedesigned in the transformed domain with orthogonal polynomials. The newlinepolynomials effects and mean square amplitude responses are computed and newlinethe features are extracted with inherent feature selection process on each newlineblock of the input scene image under consideration. The novelty of the newlineproposed feature extraction is its reduced dimensionality. The extracted newlinefeatures are then fed to the SVM classifier for the purpose of classifying the newlinescene. The proposed feature-based ISRS could produce a higher accuracy of newline83.88%. newline