Real-world scenes can be easily recognized from their edge-detected renditions: Just add motion!

Nava Rubin, Neil Albert

Research output: Contribution to journalArticle

Abstract

Background: Edge-detected version of real-world scenes are hard to recognize. At the same time, edge detection seems to play a central role in visual cortical processing. The reconciliation of this apparent discrepancy may lie in the fact that the brain operates on a dynamic, constantly changing image. Purpose: Will recognition of dynamic edge-detected real-world scenes be better than that of their static counterparts? Methods. Ten short movies (3-5 sec) of real-world scenes were recorded with a digital video camera at 30 Hz. The clips had to satisfy only one criterion: to contain motion (e.g., dogs playing, ball game). The original grayscale movies (GM) were converted to a series of still images; those were edge-detected with a standard algorithm (Canny), and re-synthesized into edge-detected movies (EDM). In addition, blank-interleaved (BI) movies were made by inserting 200 ms of blank screen after every frame of the grayscale (GBI) and edge-detected (EDBI) movies, respectively. The BI movies eliminated the sense of motion but retained the same amount of edge information. Observers were shown one of the four types of movies for each of the ten scenes, in randomized and counterbalanced order. Recognition scores were based on answers to one general ('what is the scene'?) and one specific ('how many dogs?') question. Results. Recognition of the EDMs was very high (97% correct) and not significantly different than that of GMs (98%). In contrast, recognition of the EDBIs was poor (51%), as expected from the informal observations about static edge-detected images. Good recognition of the GBIs (93%) ruled out the possibility that the poor performance in the EDBI condition resulted from any artifact of the blank-interleaving procedure. Conclusions: The information contained in edge-detected real-world scenes is more useful than previously appreciated. Neurophysiological and modeling studies need to consider how edge information is detected and processed in dynamic scenes.

Original languageEnglish (US)
JournalJournal of Vision
Volume1
Issue number3
DOIs
StatePublished - 2001

Fingerprint

Motion Pictures
Dogs
Surgical Instruments
Artifacts
Brain

ASJC Scopus subject areas

  • Ophthalmology

Cite this

Real-world scenes can be easily recognized from their edge-detected renditions : Just add motion! / Rubin, Nava; Albert, Neil.

In: Journal of Vision, Vol. 1, No. 3, 2001.

Research output: Contribution to journalArticle

@article{2bbe168c06644073b10391540d587903,
title = "Real-world scenes can be easily recognized from their edge-detected renditions: Just add motion!",
abstract = "Background: Edge-detected version of real-world scenes are hard to recognize. At the same time, edge detection seems to play a central role in visual cortical processing. The reconciliation of this apparent discrepancy may lie in the fact that the brain operates on a dynamic, constantly changing image. Purpose: Will recognition of dynamic edge-detected real-world scenes be better than that of their static counterparts? Methods. Ten short movies (3-5 sec) of real-world scenes were recorded with a digital video camera at 30 Hz. The clips had to satisfy only one criterion: to contain motion (e.g., dogs playing, ball game). The original grayscale movies (GM) were converted to a series of still images; those were edge-detected with a standard algorithm (Canny), and re-synthesized into edge-detected movies (EDM). In addition, blank-interleaved (BI) movies were made by inserting 200 ms of blank screen after every frame of the grayscale (GBI) and edge-detected (EDBI) movies, respectively. The BI movies eliminated the sense of motion but retained the same amount of edge information. Observers were shown one of the four types of movies for each of the ten scenes, in randomized and counterbalanced order. Recognition scores were based on answers to one general ('what is the scene'?) and one specific ('how many dogs?') question. Results. Recognition of the EDMs was very high (97{\%} correct) and not significantly different than that of GMs (98{\%}). In contrast, recognition of the EDBIs was poor (51{\%}), as expected from the informal observations about static edge-detected images. Good recognition of the GBIs (93{\%}) ruled out the possibility that the poor performance in the EDBI condition resulted from any artifact of the blank-interleaving procedure. Conclusions: The information contained in edge-detected real-world scenes is more useful than previously appreciated. Neurophysiological and modeling studies need to consider how edge information is detected and processed in dynamic scenes.",
author = "Nava Rubin and Neil Albert",
year = "2001",
doi = "10.1167/1.3.37",
language = "English (US)",
volume = "1",
journal = "Journal of Vision",
issn = "1534-7362",
publisher = "Association for Research in Vision and Ophthalmology Inc.",
number = "3",

}

TY - JOUR

T1 - Real-world scenes can be easily recognized from their edge-detected renditions

T2 - Just add motion!

AU - Rubin, Nava

AU - Albert, Neil

PY - 2001

Y1 - 2001

N2 - Background: Edge-detected version of real-world scenes are hard to recognize. At the same time, edge detection seems to play a central role in visual cortical processing. The reconciliation of this apparent discrepancy may lie in the fact that the brain operates on a dynamic, constantly changing image. Purpose: Will recognition of dynamic edge-detected real-world scenes be better than that of their static counterparts? Methods. Ten short movies (3-5 sec) of real-world scenes were recorded with a digital video camera at 30 Hz. The clips had to satisfy only one criterion: to contain motion (e.g., dogs playing, ball game). The original grayscale movies (GM) were converted to a series of still images; those were edge-detected with a standard algorithm (Canny), and re-synthesized into edge-detected movies (EDM). In addition, blank-interleaved (BI) movies were made by inserting 200 ms of blank screen after every frame of the grayscale (GBI) and edge-detected (EDBI) movies, respectively. The BI movies eliminated the sense of motion but retained the same amount of edge information. Observers were shown one of the four types of movies for each of the ten scenes, in randomized and counterbalanced order. Recognition scores were based on answers to one general ('what is the scene'?) and one specific ('how many dogs?') question. Results. Recognition of the EDMs was very high (97% correct) and not significantly different than that of GMs (98%). In contrast, recognition of the EDBIs was poor (51%), as expected from the informal observations about static edge-detected images. Good recognition of the GBIs (93%) ruled out the possibility that the poor performance in the EDBI condition resulted from any artifact of the blank-interleaving procedure. Conclusions: The information contained in edge-detected real-world scenes is more useful than previously appreciated. Neurophysiological and modeling studies need to consider how edge information is detected and processed in dynamic scenes.

AB - Background: Edge-detected version of real-world scenes are hard to recognize. At the same time, edge detection seems to play a central role in visual cortical processing. The reconciliation of this apparent discrepancy may lie in the fact that the brain operates on a dynamic, constantly changing image. Purpose: Will recognition of dynamic edge-detected real-world scenes be better than that of their static counterparts? Methods. Ten short movies (3-5 sec) of real-world scenes were recorded with a digital video camera at 30 Hz. The clips had to satisfy only one criterion: to contain motion (e.g., dogs playing, ball game). The original grayscale movies (GM) were converted to a series of still images; those were edge-detected with a standard algorithm (Canny), and re-synthesized into edge-detected movies (EDM). In addition, blank-interleaved (BI) movies were made by inserting 200 ms of blank screen after every frame of the grayscale (GBI) and edge-detected (EDBI) movies, respectively. The BI movies eliminated the sense of motion but retained the same amount of edge information. Observers were shown one of the four types of movies for each of the ten scenes, in randomized and counterbalanced order. Recognition scores were based on answers to one general ('what is the scene'?) and one specific ('how many dogs?') question. Results. Recognition of the EDMs was very high (97% correct) and not significantly different than that of GMs (98%). In contrast, recognition of the EDBIs was poor (51%), as expected from the informal observations about static edge-detected images. Good recognition of the GBIs (93%) ruled out the possibility that the poor performance in the EDBI condition resulted from any artifact of the blank-interleaving procedure. Conclusions: The information contained in edge-detected real-world scenes is more useful than previously appreciated. Neurophysiological and modeling studies need to consider how edge information is detected and processed in dynamic scenes.

UR - http://www.scopus.com/inward/record.url?scp=4143070474&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=4143070474&partnerID=8YFLogxK

U2 - 10.1167/1.3.37

DO - 10.1167/1.3.37

M3 - Article

VL - 1

JO - Journal of Vision

JF - Journal of Vision

SN - 1534-7362

IS - 3

ER -