Abstract:Recently, many computational models have been proposed to simulate visual cognition process. For example, the hierarchical Max-Pooling (HMAX) model was proposed according to the hierarchical and bottom-up structure of V1 to V4 in the ventral pathway of primate visual cortex, which could achieve position- and scale-tolerant recognition. In our previous work, we have introduced memory and association into the HMAX model to simulate visual cognition process. In this paper, we improve our theoretical framework by mimicking a more elaborate structure and function of the primate visual cortex. We will mainly focus on the new formation of memory and association in visual processing under different circumstances as well as preliminary cognition and active adjustment in the inferior temporal cortex, which are absent in the HMAX model. The main contributions of this paper are: 1) in the memory and association part, we apply deep convolutional neural networks to extract various episodic features of the objects since people use different features for object recognition. Moreover, to achieve a fast and robust recognition in the retrieval and association process, different types of features are stored in separated clusters and the feature binding of the same object is stimulated in a loop discharge manner and 2) in the preliminary cognition and active adjustment part, we introduce preliminary cognition to classify different types of objects since distinct neural circuits in a human brain are used for identification of various types of objects. Furthermore, active cognition adjustment of occlusion and orientation is implemented to the model to mimic the top-down effect in human cognition process. Finally, our model is evaluated on two face databases CAS-PEAL-R1 and AR. The results demonstrate that our model exhibits its efficiency on visual recognition process with much lower memory storage requirement and a better performance compared with the traditional purely computational me- hods.

A brain-inspired object-based attention network for multiobject recognition and visual reasoning

Recurrent Attention Models with Object-centric Capsule Representation for Multi-object Recognition

The attentive reconstruction of objects facilitates robust object recognition

A Brain-inspired Method for Occluded 3D Object Recognition

Learning to attend in a brain-inspired deep neural network

A Neurodynamical Cortical Model of Visual Attention and Invariant Object Recognition

Object Based Attention Through Internal Gating

Modeling Attention and Binding in the Brain through Bidirectional Recurrent Gating

A Biologically Inspired Visual Working Memory for Deep Networks

A Biologically Inspired Object-Based Visual Attention Model

Top-down attention based on object representation and incremental memory for knowledge building and inference

Rethinking Attentive Object Detection Via Neural Attention Learning.

Visual Reasoning in Object-Centric Deep Neural Networks: A Comparative Cognition Approach

Reconstruction-guided attention improves the robustness and shape processing of neural networks

The Quest for an Integrated Set of Neural Mechanisms Underlying Object Recognition in Primates

A Neurocomputational Model of Decision and Confidence in Object Recognition Task

Biologically Motivated Object Detection Based On Integration Of Top-Down And Bottom-Up Attention

Biologically Inspired Visual Model With Preliminary Cognition and Active Attention Adjustment

Enriched Deep Recurrent Visual Attention Model for Multiple Object Recognition

Brain-inspired Models for Visual Object Recognition: an Overview