Shape Feature Based Aircraft Object Recognition And Extraction From High Resolution Remotely Sensed Imagery

As the development of remote sensors, higher resolution imagery is becoming more commonly available. Therefore, object recognition and extraction of high resolution imagery has become one of the key problems in remote sensing and computer vision. As a kind of important objects, automatic aircraft object recognition and extraction plays an important role in both military and civilian fields.Compared with low resolution image, there is more spatial, geometric, textural and detailed information in high resolution image. However, the reflectance spectra is more distinct within the similar targets in high resolution image, and the traditional pixel-based method could not meet the need of some applications. For the reasons that have been mentioned, the exploration of new technology in object recognition and extraction is very necessary. Meanwhile, it should be better to mine more information, such as shape information.This paper takes aircraft as the research object to study the use of shape information in the field of object recognition and extraction in the high resolution image. An improved edge orientation histogram has been proposed to analysis the shape of the aircraft object. A recognition result can be obtained through this new shape descriptor, and the result can tell whether the aircraft objects exits, the number and the location of the aircraft objects. For several reasons, the contours of the above mentioned recognition results are often inaccurate. In order to solve the problem, a shape driven active contour model is proposed. The framework of this paper can be described as "shape feature analysis-aircraft object recognition-aircraft object extraction". The main contents and conclusions are as follows:(1) Shape feature analysisIn this part, the traditional edge orientation histogram has been improved to make it has invariance in translation, size and rotation. Experiments show that this improved shape descriptor can discriminate different objects precisely, and it can be used to recognize aircraft object.(2) Aircraft object recognitionThe aircraft object is recognized by means of windowing, that is, computer the Euclidean distance between the edge orientation histograms of current window and the aircraft sample, and if the distance is less than a certain threshold, the window may contain an aircraft object. However, experiments show this method can only tell the general location of the aircraft objects without the accurate edge information. Thus, the recognition strategy is improved, that is, the aircraft object is firstly extracted via thresholding segmentation, and then intersect it with the above results. Take the intersections as seeds and expand the segmentation results. Finally, a mathematical morphology post processing is carried out to obtain the final recognition results. The results can tell whether the aircraft objects exist, the number and the location of the aircraft objects.(3) Aircraft object extractionShape is an important visual feature and it is one of the basic features used to recognize and extract object. However, the contours of the above recognition results are often inaccurate. This is because when a 3-D real world object is projected onto a 2-D image plane, one dimension of object information is lost. Meanwhile, shape is often corrupted with noise, shadow and so on. For solving the above problem, this paper proposed a shape driven active contour which composed by three energy terms:gray term, edge term and prior shapes term. The prior shapes term restricts the deformation of the curve in a reasonable range. In order to capture the shape information, the shape samples are first matched by some methods, and then the level set method is used to represent the shape. After that, the probability distribution is defined and principle component analysis is applied to capture the variance mode of shape. Finally the shape information is combined with active contour model. The active contour model presented in this paper uses not only gray information and edge information but also prior shapes information, thus it can obtain results with high location accuracy and correct shape. Meanwhile, the recognition results are used as the initial active contour to enhance the robustness of this method.An experiment with a QuickBird image shows that, the recognition method presented in this paper can check whether the aircraft objects exist in the image. Meanwhile, it can tell the number and the location of the aircraft objects. Furthermore, the accurate contours of the aircraft objects can be recovered by means of the extraction method presented in this paper. At the same time, this paper makes quantitative evaluation on location accuracy and shape feature analysis. The location accuracy is measured by the Euclidean distance between the recognition results and the reference image, and the shape feature analysis is proceeded by computing the Hu's moment invariants.For future research, the recognition strategy should be improved to make the recognition method more effectively on large image, and the active contour model should be simplified to improve the efficiency of the algorithm.