Discriminative transform of receptive field patterns for feature representation

作者：Shu, Yucheng^*; Wang, Tianjiang; Shao, Guangpu; Hu, Chunlong

来源：Multimedia Tools and Applications, 2016, 75(13): 7495-7517.

DOI：10.1007/s11042-015-2673-7

• Receptive field
• Discrete cosine transform
• Feature representation
• Image descriptor

As reported by many neurophysiological researches, the receptive field is a basic and significant component in the human visual system. It has various kinds of properties such as orientation-selectivity, correlativity, etc. Motivated by these structural and functional properties, we propose in this paper a novel local image descriptor namely the Discriminative Transform of Receptive Field (DTRF). Specifically, around each sample pixel in the interest region, we define a low-level feature structure called Receptive Field Patterns (RFP) which is further divided into two components: the RFP-Center and RFP-Surround. Then, the local features are extracted based on Local Annular Discrete Cosine Transform (LADCT). At the descriptor construction stage, these features are pooled spatially to mimic the correlative property of receptive field. Image matching and classification experiments on four standard data set demonstrate that the proposed descriptor outperforms the state-of-the-art methods under various types of image transformations such as rotation and scaling changes, viewpoint changes, image blurring, JPEG compression, illumination changes, and image noise.

• 出版日期2016-7
• 单位江苏大学; 华中科技大学