東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

The role of local feature processing...

Stanford University.

FindBook

Google Book

Amazon

博客來

The role of local feature processing in object perception.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	The role of local feature processing in object perception./
作者:	Suh, Hyejean.
面頁冊數:	113 p.
附註:	Adviser: Kalanit Grill-Spector.
Contained By:	Dissertation Abstracts International69-02B.
標題:	Biology, Neuroscience. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3302876
ISBN:	9780549489245

The role of local feature processing in object perception.
Suh, Hyejean.

The role of local feature processing in object perception. - 113 p.

Adviser: Kalanit Grill-Spector.

Thesis (Ph.D.)--Stanford University, 2008.

One of the outstanding questions in the study of human visual perception of objects is how local feature processing affects object perception. In this thesis, we addressed this fundamental question by examining human perceptual performance on images containing whole objects or partial images of objects (containing one, two or three features) for two perceptual tasks: detection ("something" vs. random dot noise) and classification (perception of the object category, e.g., car vs. other objects). Images were embedded in visual noise and we measured human subjects' performance in a range of noise levels close to subjects' perceptual threshold. First, we found that human detection performance increases when more object area (A) is revealed and when the noise variance (N) decreases. The contribution of these factors can be characterized as a function of Ln(A/N), which we will refer to as the Area-to-Noise ratio (ANR). When the ANR was equated in the subsequent experiments, comparison of detection performance on partial vs. whole images of objects revealed the dynamic role of local feature processing in object detection: (i) detection based on a single useful feature was better than detection based on a whole object suggesting the effectiveness of local feature processing, (ii) however, detection of a whole object did not require detection of its features and (iii) detection performance varied across features; useful features (such as eyes for faces and wheels for cars) yielded better performance than suboptimal features (such as nose and mouth for faces). This pattern of results did not significantly vary across two methods we used to select features (features based on semantic judgments and a computer algorithm), and two object categories (faces and cars). For classification, results were largely similar, except that in low ANR levels (below the level sufficient for successful detection of a whole object) classification performance was better on a single useful feature than on a whole object, and in higher ANR levels the reverse was true, indicating that global processing becomes advantageous for object classification at higher ANR levels. Overall, these data suggest the dynamic role of local feature processing in object perception, which can be explained by the contribution of two mechanisms: probability summation of information in local features and spatial summation of information across whole objects.

ISBN: 9780549489245Subjects--Topical Terms:

1017680
Biology, Neuroscience.

The role of local feature processing in object perception.
LDR:03336nmm 2200289 a 45 001 867570
005 20100804
008 100804s2008 ||||||||||||||||| ||eng d
020 $a 9780549489245
035 $a (UMI)AAI3302876
035 $a AAI3302876
040 $a UMI $c UMI
100 1 $a Suh, Hyejean. $3 1036316
245 1 4 $a The role of local feature processing in object perception.
300 $a 113 p.
500 $a Adviser: Kalanit Grill-Spector.
500 $a Source: Dissertation Abstracts International, Volume: 69-02, Section: B, page: 1049.
502 $a Thesis (Ph.D.)--Stanford University, 2008.
520 $a One of the outstanding questions in the study of human visual perception of objects is how local feature processing affects object perception. In this thesis, we addressed this fundamental question by examining human perceptual performance on images containing whole objects or partial images of objects (containing one, two or three features) for two perceptual tasks: detection ("something" vs. random dot noise) and classification (perception of the object category, e.g., car vs. other objects). Images were embedded in visual noise and we measured human subjects' performance in a range of noise levels close to subjects' perceptual threshold. First, we found that human detection performance increases when more object area (A) is revealed and when the noise variance (N) decreases. The contribution of these factors can be characterized as a function of Ln(A/N), which we will refer to as the Area-to-Noise ratio (ANR). When the ANR was equated in the subsequent experiments, comparison of detection performance on partial vs. whole images of objects revealed the dynamic role of local feature processing in object detection: (i) detection based on a single useful feature was better than detection based on a whole object suggesting the effectiveness of local feature processing, (ii) however, detection of a whole object did not require detection of its features and (iii) detection performance varied across features; useful features (such as eyes for faces and wheels for cars) yielded better performance than suboptimal features (such as nose and mouth for faces). This pattern of results did not significantly vary across two methods we used to select features (features based on semantic judgments and a computer algorithm), and two object categories (faces and cars). For classification, results were largely similar, except that in low ANR levels (below the level sufficient for successful detection of a whole object) classification performance was better on a single useful feature than on a whole object, and in higher ANR levels the reverse was true, indicating that global processing becomes advantageous for object classification at higher ANR levels. Overall, these data suggest the dynamic role of local feature processing in object perception, which can be explained by the contribution of two mechanisms: probability summation of information in local features and spatial summation of information across whole objects.
590 $a School code: 0212.
650 4 $a Biology, Neuroscience. $3 1017680
650 4 $a Physics, General. $3 1018488
650 4 $a Psychology, Psychometrics. $3 1017742
690 $a 0317
690 $a 0605
690 $a 0632
710 2 $a Stanford University. $3 754827
773 0 $t Dissertation Abstracts International $g 69-02B.
790 $a 0212
790 1 0 $a Grill-Spector, Kalanit, $e advisor
791 $a Ph.D.
792 $a 2008
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3302876