Abstract In image processing field there is an attention for detection objects, regions and points then made decision in case found it in a single or collection of images may called test or image data set, for this task we have used an algorithm that used in many computer vision application and also considered very fast by compared to others this algorithm can detect and describe local features for any interest object and extract features or descriptor points from it and compare these features/ descriptor by the features that extracted from origin image, matching process has been done among features and decision made based on similar features found, this algorithm called Speeded Up Robust Features (SURF) algorithm. In this paper we used enhanced Speeded Up Robust Features "SURF" algorithm, our model counting the features in either object and origin image in data set, then matching percentage calculated using a metric of counting the size of inlier matching features towards outlier features, Radom Sample Consensus (RANSAC) algorithm has been combined with SURF for eliminated error matching that happen in features, then decision has been given based on that metric if the object is present or not. In case object found Speeded UP Robust Features "SURF" algorithm can detect the position of the interest object in origin image by using geometric transform. In this paper we have used our metric and enhanced model to made decision and write result finally for each compared process and also write some information that used for matching procedure finally we can distinguish each calculating percentage and valid strength features matched that used for finding the interest objects under different circumstances. Key words: Object detection, object recognition, feature matching, SURF.
1. Introduction The main task for object detection and recognition systems is to detect and recognize if any query object from origin image was known prior, many computer vision application recently consider the problems arises with this concept because there are problems of different images acquisition methods and the corruption of background image and noise affects in order to deal with this problems one of the robust algorithm are consider in many computer vision application called Speeded Up Robust Features "SURF" algorithm. Speeded Up Robust Features "SURF" algorithm is a local feature and descriptor algorithm that can be used in many application such as object recognition , SURF use
much larger number of features descriptor from origin image which can reduce contribution of the errors caused by local variation in the average of all feature matching . SURF can robustly recognize and identify objects in origin images even in case of clutter and partial occlusion because SURF has feature descriptor which is invariant to scale, partial variant in illumination changes and orientation. The process of Speeded UP Robust Features "SURF" algorithm can be divided into three main steps. First step is "Detection step", in this step interest points are selected at distinctive locations in the origin image, such as corners, blobs and T-junctions and this process must be robustly. The most valuable property of an interest points it s a repeatability. Repeatability express the reliability of the detector for finding the same physical interest points under different scene conditions. Second step is "Description step", in this step interest points should have unique identifiers does not depend on features scale and rotations which are called descriptor, the information of interest points represented by descriptor which are vectors that contain information about the points itself and the surroundings. Third step is "Matching step", in this step descriptor vectors are compared between the object image and the new input or origin image, the matching score is calculated based on the distance between vectors e.g. Euclidian distance and the sign of laplacian. Then if the object is found then give a message for that and view the percentage of matching score and store the result in predefined file, otherwise will give an underline message that object was not found. In this paper we are used enhanced Speeded up Robust Features "SURF" algorithm in order to detect and recognize our interest objects and proposed a metric for counting matching score to give better result. SURF algorithm was first presented by [1] in 2006, Use an integer approximation of determinant of Hessian blob detector, which can be computed with three integer operations using an integral image. SURF features descriptor are calculated by the sum of Haar wavelet response around interest points. And these can be computed by the concept of integral image. This algorithm can be used in many application such as recognize and locate of objects, track objects, face recognition, make 3D