|
Scene Classification refers to as assigning a physical scene into one of a set of predefined categories. Utilizing the
method texture feature is good for providing the approach to classify scenes. Texture can be considered to be repeating
patterns of local variation of pixel intensities. And texture analysis is important in many applications of computer image
analysis for classification or segmentation of images based on local spatial variations of intensity. Texture describes the
structural information of images, so it provides another data to classify comparing to the spectrum. Now, infrared thermal
imagers are used in different kinds of fields. Since infrared images of the objects reflect their own thermal radiation,
there are some shortcomings of infrared images: the poor contrast between the objectives and background, the effects of
blurs edges, much noise and so on. Because of these shortcomings, it is difficult to extract to the texture feature of
infrared images.
In this paper we have developed an infrared image texture feature-based algorithm to classify scenes of infrared images.
This paper researches texture extraction using Gabor wavelet transform. The transformation of Gabor has excellent
capability in analysis the frequency and direction of the partial district. Gabor wavelets is chosen for its biological
relevance and technical properties In the first place, after introducing the Gabor wavelet transform and the texture
analysis methods, the infrared images are extracted texture feature by Gabor wavelet transform. It is utilized the
multi-scale property of Gabor filter. In the second place, we take multi-dimensional means and standard deviation with
different scales and directions as texture parameters. The last stage is classification of scene texture parameters with least
squares support vector machine (LS-SVM) algorithm. SVM is based on the principle of structural risk minimization
(SRM). Compared with SVM, LS-SVM has overcome the shortcoming of higher computational burden by solving linear
equations, and has been widely used in classification and nonlinear function estimation. Some experimental results are
given in the end. The result shows that Gabor wavelet transform is successful to extract the texture feature of infrared
image. Compared with other methods the method mentioned in this paper reduces the probability of recognition and
enhances the robustness.
|
