A novel interest point detector based on convolutional features with unsupervised feature learning

Abstract

Interest point detection is of great significance in computer vision applications. This chapter presents a comparison evaluation on benchmark datasets of affine covariant features, yielding the algorithm of competitive results compared with the traditional approaches on interest point detection. Interest points, where the image gray value changes sharply, are useful low-level features that can provide informative representation for digital images. So interest point detection algorithms are the key techniques in computer vision applications such as image matching, image retrieve and 3d scene reconstruction. The chapter investigates the convolution features for interest point detection. The next one is the building of multilayer convolution network while only the first layer of convolution network is applied. The chapter presents a different detection approach based on Convolutional Neural Networks and the computation of feature maps exhibited a feature augmentation style.