Image Texture Analysis, The New Method And Its Application

Texture is a very important research work in the fields of pattern recognition and computer vision, and has an extensive application background in science research and engineering technology. Although much research work in texture analysis has been done by many researchers, and some research achievement has also been obtained, there still exist many research tasks to be resolved. The current texture analysis research at least faces the following three challenges:l)There are various texture analysis methods, but a satisfactory method in texture classification is still to be worked out, many texture researchers are attempting to explore an effective texture analysis method in a new way.2) Most of the researches to date have been done for grayscale textures although most real visual textures are in fact color textures. The research of color texture has received less attentions and an effective color texture method is to be developed.3)Many texture analysis approaches face a challenge when they are applied to other fields. The reasons for this may have been the high theory and hard implementation for the researchers and engineers from other fields.This paper mainly focuses on the three challenges of texture analysis. It first attempts to propose a novel texture analysis method in a new way, and then extends the proposed texture method to color images, and finally explores its applications in the fields of science research and engineering technology.This paper first proposes a new texture description method named Statistical Landscape Features. The proposed Statistical Landscape Features uses the information derived from the graph of an image function for texture description. The graph of an image function is a surface in the three-dimensional space that appears like a landscape. So, the problems of texture analysis are transformed into those of the analysis of the corresponding landscapes. Statistical Landscape Features first uses a variable horizontal plane to cut the image graph and derives some solids. Six texture feature curves based on the statistics of geometrical and topological properties of the solids are then used to characterize the texture. Systematic experimental comparison using the Brodatz texture set as well as the VisTex texture set shows that the performance of the proposed Statistical Landscape Features is higher than Autocorrelation Function, Edge Frequency, Fourier Transform, Spatial Grey-Level Dependance Matrix, Statistical Geometrical Features and Gabor filter. Experimental results also demonstrate that Statistical Landscape Featuresoffers high robustness to additive Gaussian noise.Most color texture analysis methods to date first derive several pseudo gray-level images from a color image, and then apply grayscale texture analysis methods to these pseudo gray-level images. The derivation of pseudo gray-level images usually regards each spectral of the color image as a pseudo gray-level image. This method does not well use hue information and may distort hue difference of the original color image. To overcome this problem, we introduce in this paper a pro-hue pseudo intensity function to derive a pseudo gray-level image from the hue component of a color image. The pro-hue pseudo intensity function of a color image can be illustrated by a gray image, which is named pro-hue pseudo intensity image, where gray values of pixels exhibit their reference color intensities. Chromatic Statistical Landscape Features derives a pseudo gray-level image from the hue component of a color image, and directly regards the saturation and intensity components as another two pseudo gray-level images. Eighteen texture feature curves are obtained from the three pseudo gray-level images when Statistical Landscape Features is separately applied to them. Chromatic Statistical Landscape Features uses these eighteen texture feature curves to describe the color texture.This paper finally explores the applications of the proposed Statistical Landscape Features in the fields of science research and engineering technology. In the field of science research, this paper considers the applications of the proposed texture analysis method in remotely-sensed image analysis and medical image analysis, and explores the applications in the browse of large remotely-sensed image database, and image classification of skin diseases and automatic detection of breast tumor regions in medical ultrasound images. In the engineering technology field, we evaluate in this paper the applications of Statistical Landscape Features on the classification and retrieval of engineering stone textures, wood textures and fabrics textures. Experimental results indicate that the proposed Statistical Landscape Features can be applied to the fields of science research and engineering technology and shows very remarkable effects.