A Hierarchical Feature Extraction Scheme with Special Vocabulary Generation for Natural Scene Classification
To automatically classify natural scenes instead of manual ways, this paper proposes a novel approach to recognize scene categories. First, we extract appearance features from an image similar to a pyramid. Then, the visual words are generated from differ
- PDF / 487,997 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 112 Downloads / 195 Views
A Hierarchical Feature Extraction Scheme with Special Vocabulary Generation for Natural Scene Classification Tian Luo, Zhuo Su and Xiaonan Luo Abstract To automatically classify natural scenes instead of manual ways, this paper proposes a novel approach to recognize scene categories. First, we extract appearance features from an image similar to a pyramid. Then, the visual words are generated from different classes separately based on Bag of Words (BOW) model. At last, Spatial Pyramid Matching (SPM) algorithm is used to obtain histogram of visual words and Support Vector Machine (SVM) is applied to classification. There are two contributions in this paper: one is that we partition an image into patches at different resolution levels and use multiple descriptors to obtain some omissive image information; the other is that visual words are formed by performing K-means clustering from each category and concatenated to form a dictionary distinguish to traditional BOW. We present satisfactory performances on a large scale of 13 categories dataset. Keywords Scene classification Support vector machine
Bag of words Spatial pyramid matching
38.1 Introduction Scene classification creates a foundation for a further recognition of objects, so it is a meaningful task to identify the semantic category an image belongs to. They can be divided into two categories in previous researches. The first one is based on T. Luo Z. Su (&) X. Luo (&) National Engineering Research Center of Digital Life, State-Province Joint Laboratory of Digital Home Interactive Applications, School of Information Science and Technology, Sun Yat-sen University, Guangzhou 510006, China e-mail: [email protected] X. Luo e-mail: [email protected]
A. A. Farag et al. (eds.), Proceedings of the 3rd International Conference on Multimedia Technology (ICMT 2013), Lecture Notes in Electrical Engineering 278, DOI: 10.1007/978-3-642-41407-7_38, Springer-Verlag Berlin Heidelberg 2014
387
388
T. Luo et al.
global features: feature extraction based on it exploits low-level pixel information to depict visual contents of an image. At present, the best way about global features regards a spatial envelope model [1] as its feature descriptor. However, these methods have high computational consumption during the feature extraction and are sensitive to image scale or brightness. Classification accuracy is also low for complex scenes. The other one is based on the ‘‘Bag of Words’’ (BOW) [2], several researchers combined this model with other new models to perform image categorization. Li et al. [3] presented a Bayesian hierarchical model to learn and recognize natural scene categories. But they did not employ spatial information among local features. The pyramid match kernel proposed by Lazebnik et al. [4] was a superior way as for the performance of matching and classification. But sharp slowdowns in performance would appear as dimension increases. Allowing for spatial relationships between two images, Grauman et al. [5] learned from Grauman’s idea and pr
Data Loading...