Image Understanding Based On Improved LBP Methodology

Image understanding is an important topic in machine vision field, it is a science that canautomatically extract from information from image, its ultimate goal is to give theinterpretation of the image, these explanations to explain the meaning of the images containedimage content.Scene classification is an important research content in Image Understandingfield. The definition of scene classification is how to find out the characteristic of the same orsimilar scenario is scene image, with multiple images in data sets, which is able to do thecorrect classifying these images, and can have the same or similar scene images for the sametype of scenario. Scene classification can achieve the goal of effective browse image libraryand make possible to retrieve or classification, this is mainly because in the process ofclassification of multiple scenarios, we can effectively organize scene images in a certain way,and in the process of scene classification can reasonable according to certain rules to completethe process. In reality, different people to the already existing same image, may also beclassified into different classes, this situation is the subjective factors are very strong, sosimple just take artificial classification method is not desirable. Scene classification can alsobe relatively easy to extract the image of a specific target.Thus,based on the background described above, and according to the specific applicationin the image retrieval, has the vital significance to understand the scene.This paper mainly introduces the basic idea of LBP operator, its origin and calculationmethods of LBP operator, and introduces the evolution and development of late LBP, such asmulti-scale LBP, rotation invariant LBP, LBP equivalent model, the calculation method ofrotation invariant equivalent model LBP and their advantage; Secondly introduces the LBPhistogram of nonparametric statistical properties, this paper expounds the application field ofLBP in the present stage, and the combination of LBP using some methods can be applied to awider area.In this paper, the main work is to study a method of image understanding explanation,and the method of classification is applied to the scene, with its content to represent the scene.This paper hope to put forward a model suitable for the actual application of spatial relations,and will further with this model the shape description, and even can be combined with imagelow-level features such as color, texture, provides a very "close" the image characteristics ofdescription, for similar image retrieval. This paper puts forward an improved LBP operator-adaptive multi-scale LBP operator, the operator used in scene image classification,, so thatcan improve the effect of image classification. Specific, in this paper, our main research contents include following some aspects: to LBP improved methods, try to merge othercharacteristics in order to achieve better experimental effect. In order to evaluate, experimentwith a variety of classifier, to find a more appropriate and a better effect experimental model.In the last experiment, we selected there famous data sets scene,which are famous inscene classification areas, eight scene categories are provided by Oliva and Torralba,15scenecategory database (scene_categories)8scenes of motion data set (event_dataset). In differentdata sets were used LBP algorithm based on the improved LBP operator to carry outexperiments in a later part of the chapter is described in detail.