Content-Driven Associative Memories for Color Image Patterns

• Published in: IEEE transactions on cybernetics Vol. 48; no. 1; pp. 139 - 150
• Main Authors: Mingming Li, Ge, Shuzhi Sam, Tong Heng Lee
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithm design and analysis; Associative memory; Associative memory (AM); Basis functions; Color; Color imagery; content-driven association; Kernel; Kernels; large-scale color images; Neural networks; Noise tolerance; Object detection; Object recognition; partial radial basis function (PRBF) kernel; Radial basis function; Random noise; Robustness; Salience; salient object detection; Tensile stress; Tensors
• ISSN: 2168-2267; 2168-2275
• DOI: 10.1109/TCYB.2016.2626500

This paper presents a novel content-driven associative memory (CDAM) to associate large-scale color images based on the subjects that represent the images' content. Compared to traditional associative memories, CDAM inherits their tolerance to random noise in images and possesses greater robustness against correlated noise that distorts an image's spatial contextual structure. A three-layer recurrent neural tensor network (RNTN) is designed as the network model of CDAM. Multiple salient objects detection algorithm and partial radial basis function (PRBF) kernel are proposed for subject determination and content-driven association, respectively. Convergence of the RNTN is analyzed based on the properties of PRBF kernels. Extensive comparative experiment results are provided to verify the CDAM's efficiency, robustness, and accuracy.