Recognition and classification of ancient dwellings based on Elastic Grid and GLCM

Yang Fan,Shen Lai-Xin

doi:10.14257/ijgdc.2016.9.5.18

Yang Fan, Shen Lai-Xin

Open Access

https://doi.org/10.14257/ijgdc.2016.9.5.18

Copy DOI

Abstract

Rectangle algorithm is designed to extract ancient dwellings from village satellite images according to their pixel features and shape features. For these unrecognized objects, we need to distinguish them by further extracting texture features of them. In order to get standardized sample, three pre-process operations including rotating operation, scaling operation, and clipping operation are designed to unify their sizes and directions. According to the partition ideas of Elastic Grid, we can choose some characteristic rows and columns of an image, and the intersecting of them form some grids, which can depict the local features of the image. Five approximate independent statistics of GLCM (Gray Level Co-occurrence Matrix) of each grid can obtain and fusion a value by using Lebesgue measure, and all the value of all grids can combine into a vector, which can reflect local features and global features of the object. We utilize classification algorithm based on GLVQ Neural Network to recognize and classify all extracted objects, and get a 3-layers classification tree. Thus, we can implement the recognition and classification of different objects and the chronology classification of different dwellings. Experiments show that the correct extracted rate of dwellings is about 76.3% by using rectangle algorithm, and the chronology classification rate of ancient dwellings is more than 88.9% by using the feature-extracting algorithm based on extended Elastic Grid and GLCM. Keyword: satellite image, GLCM, Elastic Grid, GLVQ, chronology recognition

Highlights

Huizhou area has thousands of years of ancient architecture resources that called "Old Ten" including ancient dwelling, ancient ancestral temples, ancient memorial archways, ancient trees, ancient bridges, ancient dams, ancient wells, ancient ponds, ancient water gaps and ancient pagodas [1]
According to the latitude and longitude ranges of ancient villages, the corresponding satellite images of them can be downloading from 18-levels Google Satellite Maps using the API function provided by Google Earth
There are some methods extracting global features from multimedia by using PCA [5, 14], GLCM [5, 15,16,17,18,19,20,21], GMM [15], et al In [14], the authors constructed a decision tree classification based on PCA and multi-scale texture, which can extract the types of ground objects effectively

Summary

Introduction

Huizhou area has thousands of years of ancient architecture resources that called "Old Ten" including ancient dwelling, ancient ancestral temples, ancient memorial archways, ancient trees, ancient bridges, ancient dams, ancient wells, ancient ponds, ancient water gaps and ancient pagodas [1]. There are some methods extracting global features from multimedia by using PCA [5, 14], GLCM [5, 15,16,17,18,19,20,21], GMM [15], et al In [14], the authors constructed a decision tree classification based on PCA and multi-scale texture, which can extract the types of ground objects effectively. The size of many dwellings are not same, so we need to expand the correspond size of the objects according its number of continuous pixels For these extracting objects, we use the preprocessor operators to standardize them (direction and size), and use partition technology based on improved Elastic Mesh to divide the object images and obtain their local features

Objects’ Pre-processing and Elastic Grid Partition Method

GLCM Statistics Analysis

GLVQ Classification Algorithm based on Feature Vector

Experiments and Analysis

Findings

Conclusion