On the Detection and Matching of Local Structures on Less Textured Objects Wan-Lei Zhao, Wonmin Byeon, Thomas M. Breuel ⋆ Department of Computer Science University of Kaiserslautern {zhao, byeon, tmb}@iupr.com Abstract. Due to the lack of non-zero gradients around the structures in the less textured scenes, current local feature can hardly be applied in less textured object detection. To deal with this issue, two types of local structures, namely, corner and closed region are proposed in this paper. They are based on purely object contours, which are cleaner and easier to obtain in less textured scenes. Compared to existing detec- tors, they adapt to object local structures better. In addition, these new type of local structures also bring the advantage that allows us to have different level of abstraction on the object structures. Its effectiveness has been evaluated under various image transformations and has been demonstrated with object detection in X-ray images. 1 Introduction Image local features (also known as keypoints) have been widely explored in the last decades due to its unique advantages over global features. They have been successfully applied in wide range of applications and systems, such as wide baseline matching [12], object retrieval and detection [20, 11], and near- duplicate image/video detection [20, 1]. Keypoints have been defined as the local extremas of certain measurement, which ensures their saliency and robustness to various image transformations. In general, one keypoint feature only represents one local structure in an image. It therefore has high chance of coinciding with the canonical structure of an object, which makes it possible to recognize objects by assembling their partial views. Due to the introduction of keypoint, this principle has been successfully adopted in different object detection tasks on the textured images. Many successes have been reported in different contexts about keypoint fea- tures, unfortunately most of the research about keypoint feature detection and application has been concentrating on the texture images. Although the explo- ration on keypoint feature can be partly attributed to the original search for corners in the less textured objects [21], few light has been truly shed on how ⋆ This is a technical report on recent research work that is carried out by the authors. All rights are preserved. This work is part of the SICURA project supported by Federal Ministry for Education and Research, Germany with ID FKZ 13N11125.