Аннотация:Local feature descriptors are widely employed to describe local properties of image patches when constructing a discriminative visual representation for efficient classification. Image gradient is a robust characteristic of digital image which possesses localized spatial information for each pixel along all the directions. Utilizing gradient information at the pixel level is a commonly used technique in various fields of digital image processing, especially while designing Local Feature Descriptor (LFD). In this paper, the magnitude and direction of image gradient is utilized to design an LFD. We propose a novel LFD based on Complex Gradient Function (CGF), which effectively maps a pixel from the spatial plane into its complex extension accommodating the magnitude and direction of image gradient at that pixel. To justify the genericness of the proposed descriptor, we have exploited it on two different kinds of applications, namely, human action recognition (HAR) from depth sequences and human authentication using iris biometrics. Robustness and efficiency of our descriptor are demonstrated with extensive experimental analysis on benchmark datasets IITDelhi, MMU-v2, CASIA-Iris, UBIRIS, and MICHE I for iris authentication, and MSR Action 3D dataset and MSR Gesture 3D dataset for human action recognition.