Large-Scale Scene Classification Using Gist Feature
Reza Fuad Rachmadi, I Ketut Eddy Purnama · 2014
Scene classification is one of the most challenging research problem in computer vision and image understanding areas. Vividness and viewing effect are several factors that make problem very ambiguous. In this paper, we investigate gist feature performance using several state-of-the art classifier in large-scale scene classification task. Gist feature itself is a collection of gabor filter response from image and its can represented as a region boundary of the object or shape of the scene in the image. In our experiment, we use two state-of-the art classifier, L2-regularized L2-loss SVC and SVM with RBF kernel, and SUN database to evaluating the discriminant aspects of gist feature in large-scale scene classification task. From our experiment we found that the best gist feature extraction parameters for scene classification are 18 orientations, 5 scale, and 4x4 block configuration with 16.46% accuracy in SUN database with 50 example per class for training task and SVM with RBF kernel as classifier.