A study of lacunarity-based texture analysis approaches to improve urban image classification

The traditional spectral based classifiers are normally considered ineffective in digitally classifying urban land-use and land-cover classes from high-resolution remotely sensed data due to the lack of consideration of images' spatial properties. To identify the complex arrangements of urban features in high-resolution image data, the texture information contained in a group of pixels needs to be considered. This paper discusses the concept of lacunarity and the use of two lacunarity estimation methods (i.e., binary, gray scale) in texture analysis and classification of urban images. Lacunarity has been introduced to characterize different texture appearances, which may share the same fractal dimension value. Lacunarity measures the distribution of gap sizes: low lacunarity geometric objects are homogeneous because all gap sizes are the same, whereas high lacunarity objects are heterogeneous. Using different moving windows (i.e., 13 × 13, 21 × 21, 29 × 29), the above lacunarity methods were employed to classify urban features and to observe the effects of the size of moving windows in characterizing urban texture features. Results from this study show that traditional spectral based classification approach is inaccurate in classifying urban land categories from high-resolution image data, with an accuracy of 55%, whereas lacunarity approaches can be used to improve urban classification accuracy dramatically to 92%.