A Local Feature Descriptor Based on SIFT for 3D Pollen Image Recognition

Biological particle automatic classification is an important issue in index tasking for people with pollen hypersensitivity. This paper attempts to present a local feature extraction method based on SIFT for automatic 3D pollen image recognition. In order to solve major issues in previous studies, high rate of redundant information, high feature dimensions and low recognition rate should be taken into account. Therefore, this work focuses on a four-part novel approach, including constructing 3D Gaussian pyramid to obtain muti-scale pollen images, computing the local differential vector to explore local key points, filtering the key points by inter-layer contrast, and extracting the statistical histogram descriptor of the key points as discriminant feature for automatic classification of 3D pollen images. Experiments are performed on three standard pollen image datasets including Confocal, Pollenmonitor and CHMontior. It is concluded that the descriptor can effectively describe the pollen image and is robust to the rotation, translation and scaling of the image.