Texture Descriptor And It's Application On Image Understanding

Texture is a recurring pattern on the surface of objectives.It can be used to recognize different objectives.The extracting methods of texture features can be classified as local texture features,texture features of the texture primitives and texture features by data-driven.This paper first introduces the definition,extraction methods and the applications of the three texture features.Then this paper emphatically introduces the definition,extraction methods and applications of Gabor feature and Texton feature.In the end this paper uses Gabor feature and Texton feature for grayscale image colorization respectively.Grayscale image colorization is a skill to assign color to gray images,which can be used to enhance and beautify these images.There are two categories of colorization algorithms.The first category starts with a manual input of color scribbles on the grayscale image.Then a interpolation algorithm is used to colorize the whole image.The second category colorize the grayscale image by matching with a reference color image.The algorithm proposed in this paper belongs to the second category.There are two main problems about the image colorization problem.The first one is how to make the progress automatic.The second one is how to make sure the result of the colorization is the spatial coherency during the color transfer.For the first one,this paper builds a local matching between the grayscale image and the reference color image.For the second one,this paper smoothing the coloring result with a MRF model.Experiment shows that the texture features used in this paper can benefit the grayscale image colorization.By the comparison with the method only based on the characteristics of grayscale average and variance this paper shows the importance role of texture feature in image understanding.However,there still exists some color discontinuity in the result.In order to solve this problem this paper smoothens the initial color result by using the markov random field model in post-processing.Comparison with the other colorization algorithms shows that the algorithm in this paper reaches a relatively ideal result.