A biologically inspired spiking model of visual processing for image feature detection

Abstract

Abstract:To enable fast reliable feature matching or tracking in scenes, features need to be discrete and meaningful, and hence edge or corner features, commonly called interest points are often used for this purpose. Experimental research has illustrated that biological vision systems use neuronal circuits to extract particular features such as edges or corners from visual scenes. Inspired by this biological behaviour, this paper proposes a biologically inspired spiking neural network for the purpose of image feature extraction. Standard digital images are processed and converted to spikes in a manner similar to the processing that transforms light into spikes in the retina. Using a hierarchical spiking network, various types of biologically inspired receptive fields are used to extract progressively complex image features. The performance of the network is assessed by examining the repeatability of extracted features with visual results presented using both synthetic and real images.

Bio-Inspired Ganglion Cell Models for Detecting Horizontal and Vertical Movements

“…where the parameters σ s and σ c are the standard deviations of the neighbouring and central pixels of the DoG filter [10]. The number of neurons per pixels is reduced by 2 columns and 2 rows (i.e.…”

“…Grey = (0.299 × R) + (0.587 × G) + (0.114 × B) (10) The second step is resizing the image to reduce the number of required neurons per layer; this was accomplished by using the OpenCV library built-in functions for Python 2.7. The images were scaled to 40×40 pixels while keeping the original aspect ratio.…”

“…In these works, Wu et al describes different multi-hierarchical architectures for performing image segmentation and extracting different features (including straight lines, blobs and colour features). Kerr et al [10] proposed a 4 layer hierarchical neural network for extraction complex features from natural images . Kerr et al and the Wu et al [4,5,6,7,8,9] share the following similarities: (i) the use of leaky-integrate-andfire (LIF) neuron models (see Section II-B), organized in an hierarchical network; (ii) the neural networks are composed of 3 or more layers which are interconnected via excitatory and/or inhibitory synapses, forming receptive fields; (iii) inclusion of final layer implementing spike time dependent plasticity (STDP).…”

Biologically Inspired Intensity and Depth Image Edge Extraction

“…Previous research has explored the application of SNN for artificial vision purposes. For example, segmentation, edge detection [28], contour detection [29], feature detection [14][15][16][17][18][19][20][21] and a depth from motion model [30] based on neuromorphic approaches. In [21] a hierarchical SNN is used for a visual attention system and in [30] for a categorisation system.…”

“…A Spiking Deep Belief network is used for visual classification of handwritten digits in [33] and human gesture recognition for robot partners by SNN is investigated in [34]. In [14] . Although such techniques can be readily adapted for use with depth images, none make use of the complementary information available when depth and intensity image data are combined.…”