Contour Detection-Based Discovery of Mid-Level Discriminative Patches for Scene Classification

Jinfu Yang,Jizhao Zhang,Mingai Li,Guanghui Wang

doi:10.5772/62266

Abstract

Feature extraction and representation is a key step in scene classification. In this paper, a contour detection-based mid-level features learning method is proposed for scene classification. First, a sketch tokens-based contour detection scheme is proposed to initialize seed blocks for learning mid-level patches and the patches with more contour pixels are selected as seed blocks. The procedure is demonstrated to be helpful for scene classification. Next, the seed blocks are employed to train an exemplar SVM to discover other similar occurrences and an entropy-rank criterion is utilized to mine the discriminative patches. Finally, scene categories are identified by matching the discriminative patches and testing images. Extensive experiments on the MIT Indoor-67 dataset, the 15-scene dataset and the UIUC-sports dataset show that the proposed approach yields better performance than other state-of-the-art counterparts.

Highlights

Scene classification is an important and challenging task for robots to understand the content of images in computer vision, while feature extraction and representation is a fundamental step in scene classification
We propose a new mid-level patch learning method for scene recognition based on contour detection
Inspired by [16], we introduce a contour detection algo‐ rithm based on sketch tokens in the initializing procedure of patch learning

Summary

Introduction

Scene classification is an important and challenging task for robots to understand the content of images in computer vision, while feature extraction and representation is a fundamental step in scene classification. In order to extract semantic information, high-level features based on object detection have been introduced to image representation in recent years. It has become a state-of-the-art object detection framework for multiple classes Such high-level representation has been applied to scene classification. We propose a new mid-level patch learning method for scene recognition based on contour detection. The proposed learning method with little supervision can automatically discover image patches with semantic information, which is helpful for improving scene recognition performance. 2. The proposed method of patch learning based on contour detection performs better than other counter‐ parts.

Learning Discriminative Patches

Contour detection

Definition of sketch tokens

Feature extraction and classification

UIUC-sports dataset

Discussions