Improved speeded up robust features for low contrast images

The proposed work aims at improving the feature detection in Speeded up Robust Feature (SURF) Algorithm. It has been observed that the SURF feature detector shows low feature detection in low contrast images which is caused due to the application of weighted gaussian at multiple scales before feature point detection. To overcome this problem an effective pre-processing technique is proposed which increases the image contrast to an optimum level enabling detection of more features by SURF Algorithm. The paper also introduces an effective optimization which clusters the feature points describing the same region proposal and concatenating these feature points into a single feature point with a new region proposal which holds minimal region in common with other feature points to reduce the redundant feature points generated due to the application of pre-processing. Finally, to obtain the feature vector of the new region proposal of the feature point, the feature vectors of the feature points belonging to the same cluster are concatenated to form an arbitrary dimensional feature vector.