Neural model of visual selective attention for automatic translation invariant object recognition in cluttered images

Chong, Edmund; Lim, Cheng-Chew; Lozo, P.

doi:10.1109/kes.1999.820201

Cited by 4 publications

(7 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It has been shown that Lozo's neural network can recognise familiar objects in the presence of noise 15 , however it is unable to recognise familiar images that are incomplete. Chong 16 further extends this work by proposing that in addition to top-down feedback from higher neural layers there should be a bottom-up feed-forward pathway for modulating topdown patterns. Chong 16 has shown that such an extended ART neural network is capable of recognising both partially occluded and incomplete objects in complex scenes.…”

Section: The Biological Linkmentioning

confidence: 82%

See 1 more Smart Citation

<title>Advanced correlation tracking of objects in cluttered imagery</title>

Wong

2005

SPIE Proceedings

View full text Add to dashboard Cite

Correlation tracking is used in civilian and military automatic target recognition and surveillance systems, to track objects based on their 2-dimensional shape. However traditional correlation-tracking systems have difficulty robustly detecting an object when the object is partially obscured by clutter. This paper describes one of the main problems of image-based correlation tracking systems, and proposes a novel solution. The reference image update problem occurs when the tracked object undergoes rapid shape change in the presence of clutter, here the reference image of the target is updated with an image corrupted by clutter, this can cause the system to walk-off and lose track of the target-object. The novel solution presented is based on research into modelling biological vision systems. We developed a prototype system designed to track an object changing size and shape in the presence of obscuring clutter. The system was tested on both real and simulated infrared imagery of aircraft found to be robust in the presence of obscuring clutter.

show abstract

Section: The Biological Linkmentioning

confidence: 82%

“…This includes the continuing work by Carpenter and Grossberg in the ART 11 , ART2 12 , ART3 13 and ARTmap neural networks. Of interest are the works by Lozo 14,15 and Chong 16 .…”

Section: The Biological Linkmentioning

confidence: 99%

<title>Advanced correlation tracking of objects in cluttered imagery</title>

Wong

2005

SPIE Proceedings

View full text Add to dashboard Cite

show abstract

“…Nevertheless, studies in biological neural networks have suggested that one possible function is to provide memory feedback modulation (MFM) of the bottom-up input pattern to achieve object-background separation. Inspired from these findings, the first attempt to model this functionality was the selective attention adaptive resonance theory (SAART) neural network in 1995 [28,29] and its derivatives: distortion invariant SAART in 1999 [41,42] and complementary SAART in 2001 [40,48]. The SAART neural network incorporates a mechanism for top-down memory selective attention, which is achieved via pre-synaptic amplification of the bottom-up input features by the features that are active in the shortterm memory of the neural network.…”

Section: Proposed Vision Systemmentioning

confidence: 99%

“…The SAART neural network and its extensions (DISAART and CSAART) have been demonstrated through numerous computer simulations by Lozo [28,[45][46][47], Chong [40,48,49] and Westmacott [41,42], illustrating that these networks are capable of recognising 2D shapes, when they are presented in cluttered backgrounds and under varying image distortions. However, these networks were not designed to handle large variations in the input.…”

Section: Introductionmentioning

confidence: 99%

Application of neural processing paradigm in visual landmark recognition and autonomous robot navigation

Jain

2009

Neural Comput & Applic

View full text Add to dashboard Cite

This article addresses the issue of visual landmark recognition in autonomous robot navigation along known routes, by intuitively exploiting the functions of the human visual system and its navigational ability. A feedforward-feedbackward architecture has been developed for recognising visual landmarks in real time. It integrates the theoretical concepts from the pre-attentive and attentive stages in the human visual system, the selective attention adaptive resonance theory neural network and its derivatives, and computational approaches towards object recognition in computer vision. The architecture mimics the pre-attentive and attentive stages in the context of object recognition, embedding neural network processing paradigm into a computational template-matching approach in computer vision. The real-time landmark recognition capability is achieved by mimicking the pre-attentive stage, where it models a selective attention mechanism for optimal computational resource allocation, focusing only on the regions of interest to address the computational restrictive nature of current computer processing power. Similarly, the recognition of visual landmarks in both clean and cluttered backgrounds is implemented in the attentive stage by developing a memory feedback modulation (MFM) mechanism that enables knowledge from the memory to interact and enhance the efficiency of earlier stages in the architecture. Furthermore, it also incorporates both top-down and bottom-up facilitatory and inhibition pathways between the memory and the earlier stages to enable the architecture to recognise a 2D landmark, which is partially occluded by adjacent features in the surroundings. The results show that the architecture is able to recognise objects in cluttered backgrounds using real-images in both indoor and outdoor scenes. Furthermore, the architecture application in autonomous robot navigation has been demonstrated through a number of real-time trials in both indoor and outdoor environments.

show abstract

“…the object), and (ii) subconscious execution of the program. Building on Henry James, Neisser formulated the current view of sensory "selective attention" such as auditory attention, visual attention, etc [11]. In this view, part of sensory processing is "preattentive" in that it is fast, parallel (all stimulus items simultaneously), inflexible (unchanged by behavioral goals), and involuntary (reflex-like), whereas another part is "attentive-serial" in that it is slow, serial (one stimulus item at a time), flexible (changed by behavioral goals), and voluntary.…”

Section: Biology Of Visionmentioning

confidence: 99%

<title>Techniques in biologically inspired computational vision</title>

Iftekharuddin

2005

SPIE Proceedings

View full text Add to dashboard Cite

The challenge of effectively focusing a large array of vision system in time-constrained and uncertain environments is a daunting challenge. The biology provides us some clues as higher-level organisms routinely accomplish complex tasks through iterative learning and invariance processing. In this talk, we discuss the present state-of-the-art in biological vision and recent computing models for recognition. We investigate invariance processing and reinforcement learning as some of the key ingredients for such computing. We present few implementations of these novel computing concepts.

show abstract

Neural model of visual selective attention for automatic translation invariant object recognition in cluttered images

Cited by 4 publications

References 6 publications

<title>Advanced correlation tracking of objects in cluttered imagery</title>

<title>Advanced correlation tracking of objects in cluttered imagery</title>

Application of neural processing paradigm in visual landmark recognition and autonomous robot navigation

<title>Techniques in biologically inspired computational vision</title>

Contact Info

Product

Resources

About