B-HMAX

A fast binary biologically inspired model for object recognition

Hua Zhen Zhang, Yan Feng Lu, Tae Koo Kang, Myo Taeg Lim

Research output: Contribution to journalArticle

7 Citations (Scopus)

Abstract

The biologically inspired model, Hierarchical Model and X (HMAX), has excellent performance in object categorization. It consists of four layers of computational units based on the mechanisms of the visual cortex. However, the random patch selection method in HMAX often leads to mismatch due to the extraction of redundant information, and the computational cost of recognition is expensive because of the Euclidean distance calculations for similarity in the third layer, S2. To solve these limitations, we propose a fast binary-based HMAX model (B-HMAX). In the proposed method, we detect corner-based interest points after the second layer, C1, to extract few features with better distinctiveness, use binary strings to describe the image patches extracted around detected corners, then use the Hamming distance for matching between two patches in the third layer, S2, which is much faster than Euclidean distance calculations. The experimental results demonstrate that our proposed B-HMAX model can significantly reduce the total process time by almost 80% for an image, while keeping the accuracy performance competitive with the standard HMAX.

Original languageEnglish
Pages (from-to)242-250
Number of pages9
JournalNeurocomputing
Volume218
DOIs
Publication statusPublished - 2016 Dec 19

Fingerprint

Object recognition
Information Storage and Retrieval
Visual Cortex
Costs and Cost Analysis
Hamming distance

Keywords

  • Binary descriptor
  • Classification
  • HMAX
  • Object recognition

ASJC Scopus subject areas

  • Computer Science Applications
  • Cognitive Neuroscience
  • Artificial Intelligence

Cite this

B-HMAX : A fast binary biologically inspired model for object recognition. / Zhang, Hua Zhen; Lu, Yan Feng; Kang, Tae Koo; Lim, Myo Taeg.

In: Neurocomputing, Vol. 218, 19.12.2016, p. 242-250.

Research output: Contribution to journalArticle

Zhang, Hua Zhen ; Lu, Yan Feng ; Kang, Tae Koo ; Lim, Myo Taeg. / B-HMAX : A fast binary biologically inspired model for object recognition. In: Neurocomputing. 2016 ; Vol. 218. pp. 242-250.
@article{875f57d412574e0fa92233b708c7aac0,
title = "B-HMAX: A fast binary biologically inspired model for object recognition",
abstract = "The biologically inspired model, Hierarchical Model and X (HMAX), has excellent performance in object categorization. It consists of four layers of computational units based on the mechanisms of the visual cortex. However, the random patch selection method in HMAX often leads to mismatch due to the extraction of redundant information, and the computational cost of recognition is expensive because of the Euclidean distance calculations for similarity in the third layer, S2. To solve these limitations, we propose a fast binary-based HMAX model (B-HMAX). In the proposed method, we detect corner-based interest points after the second layer, C1, to extract few features with better distinctiveness, use binary strings to describe the image patches extracted around detected corners, then use the Hamming distance for matching between two patches in the third layer, S2, which is much faster than Euclidean distance calculations. The experimental results demonstrate that our proposed B-HMAX model can significantly reduce the total process time by almost 80{\%} for an image, while keeping the accuracy performance competitive with the standard HMAX.",
keywords = "Binary descriptor, Classification, HMAX, Object recognition",
author = "Zhang, {Hua Zhen} and Lu, {Yan Feng} and Kang, {Tae Koo} and Lim, {Myo Taeg}",
year = "2016",
month = "12",
day = "19",
doi = "10.1016/j.neucom.2016.08.051",
language = "English",
volume = "218",
pages = "242--250",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",

}

TY - JOUR

T1 - B-HMAX

T2 - A fast binary biologically inspired model for object recognition

AU - Zhang, Hua Zhen

AU - Lu, Yan Feng

AU - Kang, Tae Koo

AU - Lim, Myo Taeg

PY - 2016/12/19

Y1 - 2016/12/19

N2 - The biologically inspired model, Hierarchical Model and X (HMAX), has excellent performance in object categorization. It consists of four layers of computational units based on the mechanisms of the visual cortex. However, the random patch selection method in HMAX often leads to mismatch due to the extraction of redundant information, and the computational cost of recognition is expensive because of the Euclidean distance calculations for similarity in the third layer, S2. To solve these limitations, we propose a fast binary-based HMAX model (B-HMAX). In the proposed method, we detect corner-based interest points after the second layer, C1, to extract few features with better distinctiveness, use binary strings to describe the image patches extracted around detected corners, then use the Hamming distance for matching between two patches in the third layer, S2, which is much faster than Euclidean distance calculations. The experimental results demonstrate that our proposed B-HMAX model can significantly reduce the total process time by almost 80% for an image, while keeping the accuracy performance competitive with the standard HMAX.

AB - The biologically inspired model, Hierarchical Model and X (HMAX), has excellent performance in object categorization. It consists of four layers of computational units based on the mechanisms of the visual cortex. However, the random patch selection method in HMAX often leads to mismatch due to the extraction of redundant information, and the computational cost of recognition is expensive because of the Euclidean distance calculations for similarity in the third layer, S2. To solve these limitations, we propose a fast binary-based HMAX model (B-HMAX). In the proposed method, we detect corner-based interest points after the second layer, C1, to extract few features with better distinctiveness, use binary strings to describe the image patches extracted around detected corners, then use the Hamming distance for matching between two patches in the third layer, S2, which is much faster than Euclidean distance calculations. The experimental results demonstrate that our proposed B-HMAX model can significantly reduce the total process time by almost 80% for an image, while keeping the accuracy performance competitive with the standard HMAX.

KW - Binary descriptor

KW - Classification

KW - HMAX

KW - Object recognition

UR - http://www.scopus.com/inward/record.url?scp=84994130278&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84994130278&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2016.08.051

DO - 10.1016/j.neucom.2016.08.051

M3 - Article

VL - 218

SP - 242

EP - 250

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

ER -