Facial identification in very low-resolution images simulating prosthetic vision

M. H. Chang, H. S. Kim, J. H. Shin, Kwangsuk Park

Research output: Contribution to journalArticleResearchpeer-review

16 Citations (Scopus)

Abstract

Familiar facial identification is important to blind or visually impaired patients and can be achieved using a retinal prosthesis. Nevertheless, there are limitations in delivering the facial images with a resolution sufficient to distinguish facial features, such as eyes and nose, through multichannel electrode arrays used in current visual prostheses. This study verifies the feasibility of familiar facial identification under low-resolution prosthetic vision and proposes an edge-enhancement method to deliver more visual information that is of higher quality. We first generated a contrast-enhanced image and an edge image by applying the Sobel edge detector and blocked each of them by averaging. Then, we subtracted the blocked edge image from the blocked contrast-enhanced image and produced a pixelized image imitating an array of phosphenes. Before subtraction, every gray value of the edge images was weighted as 50% (mode 2), 75% (mode 3) and 100% (mode 4). In mode 1, the facial image was blocked and pixelized with no further processing. The most successful identification was achieved with mode 3 at every resolution in terms of identification index, which covers both accuracy and correct response time. We also found that the subjects recognized a distinctive face especially more accurately and faster than the other given facial images even under low-resolution prosthetic vision. Every subject could identify familiar faces even in very low-resolution images. And the proposed edge-enhancement method seemed to contribute to intermediate-stage visual prostheses.

Original languageEnglish
Article number046012
JournalJournal of Neural Engineering
Volume9
Issue number4
DOIs
StatePublished - 1 Aug 2012

Fingerprint

Visual Prosthesis
Image resolution
Prosthetics
Phosphenes
Feasibility Studies
Nose
Reaction Time
Electrodes
Detectors
Processing

Cite this

@article{611ffcc33d46499a88d91322b3efb2ae,
title = "Facial identification in very low-resolution images simulating prosthetic vision",
abstract = "Familiar facial identification is important to blind or visually impaired patients and can be achieved using a retinal prosthesis. Nevertheless, there are limitations in delivering the facial images with a resolution sufficient to distinguish facial features, such as eyes and nose, through multichannel electrode arrays used in current visual prostheses. This study verifies the feasibility of familiar facial identification under low-resolution prosthetic vision and proposes an edge-enhancement method to deliver more visual information that is of higher quality. We first generated a contrast-enhanced image and an edge image by applying the Sobel edge detector and blocked each of them by averaging. Then, we subtracted the blocked edge image from the blocked contrast-enhanced image and produced a pixelized image imitating an array of phosphenes. Before subtraction, every gray value of the edge images was weighted as 50{\%} (mode 2), 75{\%} (mode 3) and 100{\%} (mode 4). In mode 1, the facial image was blocked and pixelized with no further processing. The most successful identification was achieved with mode 3 at every resolution in terms of identification index, which covers both accuracy and correct response time. We also found that the subjects recognized a distinctive face especially more accurately and faster than the other given facial images even under low-resolution prosthetic vision. Every subject could identify familiar faces even in very low-resolution images. And the proposed edge-enhancement method seemed to contribute to intermediate-stage visual prostheses.",
author = "Chang, {M. H.} and Kim, {H. S.} and Shin, {J. H.} and Kwangsuk Park",
year = "2012",
month = "8",
day = "1",
doi = "10.1088/1741-2560/9/4/046012",
language = "English",
volume = "9",
journal = "Journal of Neural Engineering",
issn = "1741-2560",
publisher = "IOP Publishing Ltd.",
number = "4",

}

Facial identification in very low-resolution images simulating prosthetic vision. / Chang, M. H.; Kim, H. S.; Shin, J. H.; Park, Kwangsuk.

In: Journal of Neural Engineering, Vol. 9, No. 4, 046012, 01.08.2012.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Facial identification in very low-resolution images simulating prosthetic vision

AU - Chang, M. H.

AU - Kim, H. S.

AU - Shin, J. H.

AU - Park, Kwangsuk

PY - 2012/8/1

Y1 - 2012/8/1

N2 - Familiar facial identification is important to blind or visually impaired patients and can be achieved using a retinal prosthesis. Nevertheless, there are limitations in delivering the facial images with a resolution sufficient to distinguish facial features, such as eyes and nose, through multichannel electrode arrays used in current visual prostheses. This study verifies the feasibility of familiar facial identification under low-resolution prosthetic vision and proposes an edge-enhancement method to deliver more visual information that is of higher quality. We first generated a contrast-enhanced image and an edge image by applying the Sobel edge detector and blocked each of them by averaging. Then, we subtracted the blocked edge image from the blocked contrast-enhanced image and produced a pixelized image imitating an array of phosphenes. Before subtraction, every gray value of the edge images was weighted as 50% (mode 2), 75% (mode 3) and 100% (mode 4). In mode 1, the facial image was blocked and pixelized with no further processing. The most successful identification was achieved with mode 3 at every resolution in terms of identification index, which covers both accuracy and correct response time. We also found that the subjects recognized a distinctive face especially more accurately and faster than the other given facial images even under low-resolution prosthetic vision. Every subject could identify familiar faces even in very low-resolution images. And the proposed edge-enhancement method seemed to contribute to intermediate-stage visual prostheses.

AB - Familiar facial identification is important to blind or visually impaired patients and can be achieved using a retinal prosthesis. Nevertheless, there are limitations in delivering the facial images with a resolution sufficient to distinguish facial features, such as eyes and nose, through multichannel electrode arrays used in current visual prostheses. This study verifies the feasibility of familiar facial identification under low-resolution prosthetic vision and proposes an edge-enhancement method to deliver more visual information that is of higher quality. We first generated a contrast-enhanced image and an edge image by applying the Sobel edge detector and blocked each of them by averaging. Then, we subtracted the blocked edge image from the blocked contrast-enhanced image and produced a pixelized image imitating an array of phosphenes. Before subtraction, every gray value of the edge images was weighted as 50% (mode 2), 75% (mode 3) and 100% (mode 4). In mode 1, the facial image was blocked and pixelized with no further processing. The most successful identification was achieved with mode 3 at every resolution in terms of identification index, which covers both accuracy and correct response time. We also found that the subjects recognized a distinctive face especially more accurately and faster than the other given facial images even under low-resolution prosthetic vision. Every subject could identify familiar faces even in very low-resolution images. And the proposed edge-enhancement method seemed to contribute to intermediate-stage visual prostheses.

UR - http://www.scopus.com/inward/record.url?scp=84863827077&partnerID=8YFLogxK

U2 - 10.1088/1741-2560/9/4/046012

DO - 10.1088/1741-2560/9/4/046012

M3 - Article

VL - 9

JO - Journal of Neural Engineering

JF - Journal of Neural Engineering

SN - 1741-2560

IS - 4

M1 - 046012

ER -