Line Matching Method Based On A New Geometric Invariant:CHR-IMP Method

As feature matching technology is used in many applications, feature matching attracts more and more attention. Feature matching aims to recover the correspondence of features including points, curves/lines and regions projected into multiple imaging planes (views). Up to now, research on point matching is more and more deep and mature. However, most point matching methods are mainly based on image textural information. In practice, there are numerous scenarios not including rich textural information, especially for man-made scenes (sequence image) such as images for buildings and indoor furniture. Thereafter, they are not able to obtain enough feature points. Besides, owning to the scarcity of recent region matching methods as well as the irregularity of region features, there are many enormous difficulties in region matching. For the reasons above, line feature matching should comes first, and it becomes a fundamental task in many practical applications for these man-made scenes (sequence image).This paper comes up with a robust line matching method based on Characteristic Ratio (CHR) invariant of collinear points-CHR-IMP method, without resorting to any other constraints or priori knowledge. Main contents are as follows:(1).This paper introduces a new affine invariant named as Characteristic Ratio (CHR), which is an. Since it reflects the geometric relationship between any number of collinear points, it is possible to choose any number of collinear points flexibly (here number is3) to calculate characteristic ratio with simple value, which is of high efficiency;(2). In order to effectively calculating the variant CHR used to line matching, we define a set of collinear points, denoted as Important Point (IMP). Its definition embody an important idea:In some research fields, such as image matching, we should try our best to use a small amount of important information which can fully reflect the characteristic of images. IMPs are capable of reliably computing the values of CHR and have high efficiency. Besides, the IMPs across views are matched by a similarity metric named the nearest neighbor distance ratio (NNDR).(3). When matching the lines IMPs lie, we simply use an existing invariant, i.e., cross product symbol familiar to everyone. In fact, an important geometric property of it is that it can reflect the relative clockwise or counter-clockwise relationship between two vectors. Thus, we apply this invariant on our line matching process based on this property, and show the specific algorithm;(4). We compare and analysis our approach with the other latest line matching algorithms under various cases. The experiment results fully show that CHR-IMP method can deal with some line matching problem under those cases (scaling, occlusion) that other methods couldn’t solve. Besides, this method has high operation efficiency and robust results.