๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ
Study/Human-Computer Interaction

[JOV-2007] Measuring visual clutter(๋…ผ๋ฌธ๋ฆฌ๋ทฐ)

by ์œ ๋ฏธ๋ฏธYoomimi 2023. 8. 2.

 

์˜ค๋Š˜ ์š”์•ฝํ•ด๋ณผ ๋…ผ๋ฌธ์€ ์ตœ๊ทผ ์ฝ์€

Ruth Rosenholtz, Yuanzhen Li, Lisa Nakano, "Measuring visual clutter", Journal of Vision August 2007, Vol.7, 17. ์ž…๋‹ˆ๋‹ค.

 

[JOV-2007]Measuring visual clutter https://doi.org/10.1167/7.2.17

 

Measuring visual clutter | JOV | ARVO Journals

We again looked at RTs for correct trials, separating target-present from target-absent trials. For each image, we averaged RT over Gabor type, target location, and subject. We found that mean RT was significantly slower in the original images than in the

jov.arvojournals.org

 

 

*๋ชจ๋“  ์ด๋ฏธ์ง€ ์ž๋ฃŒ๋Š” ํ•ด๋‹น ๋…ผ๋ฌธ์—์„œ ๊ฐ€์ ธ์™”์Šต๋‹ˆ๋‹ค.

 


 

Visual Clutter๋ž€?

 

 Visual Clutter๋Š” ๋ณต์žกํ•œ ์ด๋ฏธ์ง€์—์„œ์˜ Visual research Difficulty์— ๋Œ€ํ•œ ์ดํ•ด๋ฅผ ๋•๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉ๋˜๋Š” ์ผ์ข…์˜ ์‹œ๊ฐ ํ˜ผ์žก๋„์ง€ํ‘œ์ด๋‹ค. ์ด Paper์—์„œ๋Š” Visual Clutter๋ฅผ, ๋‹จ์ˆœํžˆ item(object)์˜ ์ˆ˜๊ฐ€ ์•„๋‹Œ, item์˜ ์ถ”๊ฐ€๊ฐ€ search performance์˜ ์ €ํ•˜๋ฅผ ์ดˆ๋ž˜ํ•˜๋Š” ์ •๋„๋ผ๊ณ  ์ •์˜ํ•˜๊ณ  ์žˆ๋‹ค. (์˜ˆ์™ธ์ ์œผ๋กœ item์ด ๋งŽ์„ ๋•Œ search performance๊ฐ€ ํ–ฅ์ƒ๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ๋‹ค. item๋“ค์— pattern์ด ์žˆ๋Š” ๊ฒฝ์šฐ์— ๊ทธ๋Ÿฌํ•˜๋‹ค.)

 

 

์ด ๋…ผ๋ฌธ์—์„œ ์ œ์•ˆํ•˜๋Š” ์„ธ ๊ฐ€์ง€ ์ธก์ •๋ฒ•์„ ๊ฐ„๋‹จํžˆ ์š”์•ฝํ•˜์ž๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 

"Feature Congestion”: Statistical saliency model(์ƒˆ๋กœ์šด item์„ ์ถ”๊ฐ€ํ•œ๋‹ค๊ณ  ๊ฐ€์ •ํ–ˆ์„ ๋•Œ ๋ˆˆ์— ๋Œ ์ˆ˜ ์žˆ๋Š” ์ •๋„)์„ ์ด์šฉํ•ด color, orientation์— ๋Œ€ํ•œ ๊ฐ๊ฐ์˜ clutter์„ ๊ตฌํ•˜๊ณ  ์—ฌ๊ธฐ์— contrast clutter(๋‹จ์ˆœํžˆ variance)๋ฅผ ๋”ฐ๋กœ ๊ตฌํ•ด ํ•ฉํ•œ๋‹ค.

"Subband Entropy": Shannon entropy(์ด๋ฏธ์ง€๋ฅผ encodingํ•˜๊ธฐ ์–ด๋ ค์šด ์ •๋„)๋ฅผ ์ด์šฉํ•ด luminance์™€ chrominance ์ฑ„๋„์—์„œ ๊ฐ๊ฐ์˜ clutter๋ฅผ ๊ตฌํ•ด ํ•ฉํ•œ๋‹ค.

"Edge Density": ์ด๋ฏธ์ง€์— ๋‹ด๊ธด object๋“ค์˜ edge๋ฅผ ์ธก์ •ํ•ด edge pixels๊ฐ€ ์ „์ฒด pixels์—์„œ ์ฐจ์ง€ํ•˜๋Š” percentage๋ฅผ ๊ตฌํ•œ๋‹ค.

 

ํ•˜๋‚˜์”ฉ ์ž์„ธํžˆ ์•Œ์•„๋ณด์ž.

 


 

Feature Congestion

 

 Feature Congestion ์ธก์ •๋ฐฉ์‹์— ๋Œ€ํ•ด ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด, ๋จผ์ € The statistical saliency model์— ๋Œ€ํ•ด ์ดํ•ดํ•˜์—ฌ์•ผ ํ•œ๋‹ค.

‘The Statistical Saliency Model’์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 ์—ฌ๊ธฐ์„œ T๋Š” local distribution of feature vectors์— ๋Œ€ํ•œ ์ด์ƒ์น˜๋ฅผ ๋œปํ•œ๋‹ค. μd๋Š” T์˜ mean, d๋Š” T๋“ค์˜ covariance์ด๋‹ค. ์ด๋ ‡๊ฒŒ ๊ณ„์‚ฐ๋œ delta ๊ฐ’์€ saliency๋ฅผ ๋œปํ•œ๋‹ค.

 Saliency model์€ Mean๊ณผ Covariance์— ์˜ํ•ด ์™ผ์ชฝ๊ณผ ๊ฐ™์€ ellipsoids๋กœ ํ‘œํ˜„๋  ์ˆ˜ ์žˆ๋‹ค. (๊ฐ€์žฅ ์•ˆ์ชฝ ํƒ€์›์€ mean feature vector์™€ standard deviation = 1๋งŒํผ ๋–จ์–ด์ ธ ์žˆ๋‹ค๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•˜๊ณ , ๊ทธ ๋‹ค์Œ์€ 2, 3, 4๋งŒํผ ๋–จ์–ด์ ธ ์žˆ๋Š” ํƒ€์› ๊ทธ๋ฆผ์ด๋‹ค.) ์ด ellipsoid์™€ target vector๊ฐ€ ์–ผ๋งˆ๋‚˜ ๊ฐ€๊นŒ์šด์ง€๋ฅผ ํ†ตํ•ด target searching์ด ์–ผ๋งˆ๋‚˜ ์–ด๋ ค์šด์ง€ ํŒ๋‹จํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋ฆผ์—์„œ ๊ฒ€์€ ๋™๊ทธ๋ผ๋ฏธ๊ฐ€ ํฐ ๋™๊ทธ๋ผ๋ฏธ๋ณด๋‹ค ํƒ€์›์œผ๋กœ๋ถ€ํ„ฐ ๋” ๊ฐ€๊นŒ์ด ์žˆ์œผ๋ฏ€๋กœ, ๋น„๊ต์  searching์ด ์–ด๋ ค์šธ ๊ฒƒ์ด๋ž€ ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค.

 

 ๋”ฐ๋ผ์„œ saliency ํƒ€์›์ฒด์˜ volume์ด ํฌ๋‹ค๋ฉด ์ƒˆ๋กœ์šด ์š”์†Œ๋ฅผ ์ถ”๊ฐ€ํ•ด๋„ ๋ˆˆ์— ๋„์ง€ ์•Š์„ ํ™•๋ฅ ์ด ์ปค์ง„๋‹ค๋Š” ๋œป์ด ๋˜๋ฏ€๋กœ, ํƒ€์›์ฒด์˜ volume์ด ์ปค์ง€๋ฉด clutter๊ฐ€ ์ปค์ง„๋‹ค๊ณ  ์ƒ๊ฐํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๋ฐ์—์„œ clutter ์ธก์ •๋ฒ•์„ ์ฐฉ์•ˆํ–ˆ๋‹ค.

 

 ์ด๋“ค์€ ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•ด color, orientation, luminance contrast์„ ๊ณ ๋ คํ•˜์—ฌ clutter์„ ์ธก์ •ํ•˜๋Š” ๋ฒ•์„ ์ƒˆ๋กญ๊ฒŒ ๊ณ ์•ˆํ•˜๊ณ , ์ด๋ฅผ Feature Congestion์ด๋ผ๊ณ  ํ•˜์˜€๋‹ค. ๊ทธ ๋ฐฉ์‹์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 

1) ์ด๋ฏธ์ง€์˜ ์—ฌ๋Ÿฌ part(local image)์— ๋Œ€ํ•ด feature๋งˆ๋‹ค, scale๋งˆ๋‹ค (co)variance๋ฅผ ๊ตฌํ•œ๋‹ค.

 CIELab color space์— ๊ธฐ๋ฐ˜ํ•ด, Gaussian pyramid์— ๋”ฐ๋ผ ์ด๋ฏธ์ง€๋ฅผ 3๊ฐœ์˜ scale๋กœ ๋‚˜๋ˆˆ๋‹ค. ์ด๋ ‡๊ฒŒ ๋‚˜๋‰œ ๊ฐ scale๋งˆ๋‹ค features(color, orientation, contrast)๋ณ„๋กœ vector๋ฅผ ์ฐพ๋Š”๋‹ค. ์ด๋•Œ Color์€ Gaussian filter๋กœ poolingํ•ด local mean color๋ฅผ ์ถ”์ถœํ•œ ๋’ค ์ด ๊ฐ’์˜ Covariance ellipsoid volume์„ ๊ตฌํ•˜๊ณ , Orientation์€ scale๋งˆ๋‹ค two-vector๋ฅผ ์–ป์–ด๋‚ด์„œ ๊ตฌํ•œ oriented opponent energy ๊ฐ’์˜ Covariance ellipsoid volume์„ ๊ตฌํ•œ๋‹ค. Luminance contrast๋Š” center-surround filter๋กœ luminance band๋ฅผ filteringํ•œ ๊ฒฐ๊ณผ๋ฅผ ์ œ๊ณฑํ•˜์—ฌ variance๋ฅผ ๊ตฌํ•œ๋‹ค.

 

2) scale์— ๋”ฐ๋ผ clutter๋ฅผ ํ•ฉ์ณ feature type๋ณ„๋กœ clutter๋ฅผ ๊ตฌํ•œ๋‹ค.

 3๊ฐœ์˜ scale ์ค‘ ํ•œ ๊ฐ€์ง€ scale์—์„œ๋งŒ cluttered ๋˜์–ด๋„ ๊ทธ feature์—์„œ cluttered๋ผ๊ณ  ๋ด์•ผ ํ•œ๋‹ค๋Š” ๊ฐ€์ • ํ•˜์—, ๊ฐ๊ฐ์˜ pixel์—์„œ ๊ฐ feature๋งˆ๋‹ค scale์— ๋”ฐ๋ฅธ clutter ๊ฐ’ ์ค‘ ์ตœ๋Œ€๊ฐ’์„ ๊ทธ pixel์—์„œ ํ•ด๋‹น feature์˜ clutter ๊ฐ’์œผ๋กœ ์ทจํ•œ๋‹ค.

 

3) pixel ๋งˆ๋‹ค feature type ๋ณ„ clutter๋ฅผ ํ•˜๋‚˜๋กœ ํ•ฉ์นœ๋‹ค.

Color ๊ฐ’์—๋Š” ์„ธ์ œ๊ณฑ๊ทผ์„, orient ๊ฐ’์—๋Š” ์ œ๊ณฑ๊ทผ์„ ์ทจํ•˜์—ฌ contrast ๊ฐ’๊ณผ์˜ scale์„ ๋งž์ถฐ์ฃผ๊ณ , ๊ฐ๊ฐ ์ •๊ทœํ™” ํ•œ ๋‹ค์Œ, pixel๋งˆ๋‹ค ์„ธ ๊ฐ€์ง€ feature clutter์˜ linear sum์„ ๊ตฌํ•œ๋‹ค.

 

4) ์ „์ฒด ์ด๋ฏธ์ง€์— ๋Œ€ํ•ด ํ•˜๋‚˜์˜ clutter ๊ฐ’์„ ๊ตฌํ•œ๋‹ค.

 pixel clutter์˜ ํ‰๊ท ์„ ๊ตฌํ•ด ๊ทธ ๊ฐ’์„ ์ด๋ฏธ์ง€ ์ „์ฒด์˜ clutter (measured by Feature Congestion) ๊ฐ’์œผ๋กœ ์ทจํ•œ๋‹ค.

 

 

 

 

Subband Entropy

 

 Clutter๋ฅผ์ด๋ฏธ์ง€๊ฐ€ ์–ผ๋งˆ๋‚˜ organized ๋˜์–ด์žˆ๋Š”์ง€’, ๋‚˜์•„๊ฐ€ ‘encoding์„ ์œ„ํ•ด ์–ผ๋งŒํผ์˜ ๋…ธ๋ ฅ(bits)์ด ํ•„์š”ํ•œ์ง€๋กœ ํŒ๋‹จํ•  ์ˆ˜๋„ ์žˆ๋‹ค. clutter๊ฐ€ ๋‚ฎ์œผ๋ฉด ์ด๋ฏธ์ง€์˜ ์ค‘๋ณต์„ฑ์ด ํฌ๋‹ค๋Š” ๊ฒƒ์ด๊ณ , ์ด๋Š” ๋” ํšจ์œจ์ ์œผ๋กœ encodingํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๋œป์ด ๋œ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด clutter๋ฅผ ์ธก์ •ํ•˜๋Š” ๋‘๋ฒˆ์งธ ๋ฐฉ๋ฒ•์œผ๋กœ subband entropy๋ฅผ ์ด์šฉํ•œ ์ธก์ •์„ ์ œ์•ˆํ•œ๋‹ค.

 

 ์ด ์ธก์ •๋ฒ•์€ ‘subband(wavelet) image coding์„ ํ•˜๊ธฐ ์œ„ํ•ด ์š”๊ตฌ๋˜๋Š” ์–ด๋ ค์›€์˜ ์ •๋„๊ฐ€ clutter๋ฅผ ๊ฒฐ์ •ํ•œ๋‹ค.’๊ณ  ๊ฐ€์ •ํ•˜๊ณ  ์ด๋ฅผ ์ด์šฉํ•ด clutter๋ฅผ ์ˆ˜์น˜ํ™” ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ๋งŽ์€ image encoder๋“ค์ด entropy encoding์„ ์ด์šฉํ•ด ์ด๋ฏธ์ง€๋ฅผ ๋ณ€ํ™˜ํ•œ๋‹ค๋Š” ๋ฐ์—์„œ ์ฐฉ์•ˆํ•˜์—ฌ, subband entropy measure์—์„œ๋Š” ์ด๋ฏธ์ง€์—์„œ ํŠน์ • subband ๋งˆ๋‹ค Shannon entropy๋ฅผ ๊ตฌํ•ด ์ด์šฉํ•œ๋‹ค.

 Shannon entropy ๊ณต์‹์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 ์—ฌ๊ธฐ์„œ p๋Š” the probability distribution of coefficients in each subband์ด๋‹ค. ์ด ๊ฐ’์€ ๊ฐ subband ์—์„œ์˜ coefficient๋ฅผ binningํ•˜์—ฌ ์ธก์ •ํ•œ๋‹ค. ์ด๋•Œ ๋” ์„ธ๋ฐ€ํ•œ interval์ด ์š”๊ตฌ๋œ๋‹ค๋Š” ๊ฒƒ์€ encoding์ด ๋” ์–ด๋ ค์›Œ์ง„๋‹ค๋Š” ์˜๋ฏธ๊ฐ€ ๋œ๋‹ค. ๊ตฌ์ฒด์ ์ธ clutter ์ธก์ • ๋ฐฉ์‹์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 

 

1) CIELab์„ ์ด์šฉํ•ด ์ด๋ฏธ์ง€๋ฅผ convertํ•˜์—ฌ ์ด๋ฏธ์ง€์˜ RGB ๊ฐ’์„ ์–ป์–ด๋‚ธ๋‹ค.

 

2) steerable pyramid๋ฅผ ์ด์šฉํ•ด ์ด๋ฏธ์ง€ RGB ์ •๋ณด๋ฅผ luminance(L)(ํœ˜๋„)์™€ chrominance(a, b)(์ƒ‰์ฐจ) ์ •๋ณด๋ฅผ ๊ฐ–๋Š” wavelet subband๋กœ ๋ถ„ํ•ดํ•œ๋‹ค.

 

3) Shannon entropy ๊ณต์‹์„ ์ด์šฉํ•ด ๊ฐ๊ฐ์˜ subband์—์„œ wavelet coefficient๋ฅผ ์ธก์ •ํ•œ๋‹ค.

 

4) Luminance์™€ Chrominance channel ๊ฐ๊ฐ์—์„œ subband entropies๋ฅผ ํ•ฉํ•˜์—ฌ ํ‰๊ท ์„ ๊ตฌํ•ด ๊ฐ ์ฑ„๋„์—์„œ์˜ subband entropy๋ฅผ ๊ตฌํ•œ๋‹ค.

 

5) ๊ฐ ์ฑ„๋„์˜ subband entropies๋ฅผ scaling ํ•˜์—ฌ ํ•ฉํ•œ๋‹ค.

 

 

 

 

Edge Density

 

 ์„ ํ–‰ ์—ฐ๊ตฌ์— ๋”ฐ๋ฅด๋ฉด, ์‚ฌ๋žŒ์ด ๋ณต์žกํ•จ์„ ๋Š๋ผ๊ฒŒ ํ•˜๋Š” ์š”์ธ์—๋Š” object์˜ ์ˆ˜, detail, color ๋“ฑ์˜ ๋‹ค์–‘ํ•œ ์š”์ธ๋“ค ์™ธ์—๋„ ‘openness’๊ฐ€ ์žˆ๋‹ค. ์ด๋Š” ‘Edge density’๋ผ๋Š” ๊ฐœ๋…์œผ๋กœ ์ธก์ •๋  ์ˆ˜ ์žˆ๋Š”๋ฐ, ์—ฌ๊ธฐ์„œ edge๋Š” ์ด๋ฏธ์ง€ ์•ˆ์˜ object๋“ค์˜ edge๋ฅผ ๋งํ•˜๊ณ , Edge density๋Š” ์ „์ฒด์—์„œ edge pixels์ด ์ฐจ์ง€ํ•˜๋Š” ๋น„์œจ์„ ๋œปํ•œ๋‹ค. ์ด ์—ฐ๊ตฌ์—์„œ๋Š” Canny edge detector๋ฅผ ์ด์šฉํ•ด edge pixels์˜ density๋ฅผ ์ธก์ •ํ–ˆ๋‹ค.

 

 


 

Evaluation of measures of visual clutter

 

 ์ด๋ ‡๊ฒŒ 3๊ฐ€์ง€ ๋ฐฉ์‹์œผ๋กœ clutter๋ฅผ ์ธก์ •ํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํ•  ๋•Œ, ํŠน์ • ์ƒํ™ฉ์—์„œ ์–ด๋–ค ์ธก์ •๋ฒ•์ด ์ ์ ˆํ•œ์ง€๋ฅผ, Clutter๊ฐ€ ์ธ๊ฐ„์˜ ์ธ์ง€ ๋Šฅ๋ ฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ํ† ๋Œ€๋กœ ์‹คํ—˜ํ•˜์˜€๋‹ค.

 ์‹คํ—˜์€ ํฌ๊ฒŒ 3๊ฐ€์ง€๋‹ค. ์ฒซ๋ฒˆ์งธ๋Š” ‘reaction time in visual search according to clutter’, ๋‘๋ฒˆ์งธ๋Š” ‘contrast thresholds in visual search according to clutter’, ๋งˆ์ง€๋ง‰์€ ‘whether color variability matters’์— ๊ด€ํ•œ ์‹คํ—˜์ด๋‹ค.

 

 

"Experiment 1: Visual search reaction time in cluttered maps"๋Š” visual search ์ƒํ™ฉ์—์„œ clutter๊ฐ€ reaction time์— ์–ด๋–ค ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š”์ง€์— ๊ด€ํ•œ ์‹คํ—˜์ด๋‹ค. ์ฒ˜์Œ์— ์–ธ๊ธ‰ํ–ˆ๋˜ ๋Œ€๋กœ ์ด paper์—์„œ๋Š” clutter๋ฅผ ‘search performance๋ฅผ ์ €ํ•˜์‹œํ‚ค๋Š” ์ƒํƒœ๋กœ ์ •์˜ํ•˜๊ณ  ์žˆ์œผ๋ฏ€๋กœ, ์ œ์•ˆ๋œ ์ธก์ •๋ฒ•์œผ๋กœ ์ธก์ •ํ•œ clutter ๊ฐ’์ด ์ปค์ง์— ๋”ฐ๋ผ reaction time์ด ์ปค์ง€๋ฉด clutter ์ธก์ •๋ฒ•์ด ์ ์ ˆํ–ˆ๋‹ค๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค.

 

 ์ด ์‹คํ—˜์—์„œ๋Š” ์ฐธ๊ฐ€์ž๋“ค์ด ์ง€๋„์—์„œ ํŠน์ • target(gabor)๋ฅผ ์ฐพ๋Š” ์ž‘์—…์„ ์ˆ˜ํ–‰ํ–ˆ๋‹ค. ํ•ด๋‹น ์ด๋ฏธ์ง€์— ๋ชฉํ‘œ target์ด present๋ฉด “f”, absent๋ฉด “j”๋ฅผ ๋ˆ„๋ฅด๊ฒŒ ํ–ˆ๋‹ค. ๋ฒ„ํŠผ์„ ๋ˆ„๋ฅด๊ธฐ๊นŒ์ง€ ๊ฑธ๋ฆฐ ์‹œ๊ฐ„์„ ์ธก์ •ํ–ˆ๋‹ค. Background ์ด๋ฏธ์ง€์˜ ์ข…๋ฅ˜๋Š” clutter์— ๋”ฐ๋ผ 19๊ฐ€์ง€, target์˜ ์ข…๋ฅ˜๋Š” 16๊ฐ€์ง€(๊ฐ๊ธฐ ๋‹ค๋ฅธ grayscale gabor), target์ด ๋‚˜ํƒ€๋‚  ์ˆ˜ ์žˆ๋Š” ์œ„์น˜๋Š” 6๊ตฐ๋ฐ์— ๋”ฐ๋ผ target์˜ ์กด์žฌ ์—ฌ๋ถ€(2๊ฐ€์ง€)๊นŒ์ง€ ๋‚˜๋‰˜์–ด ์ด 3,648 ๊ฐœ์˜ ์ด๋ฏธ์ง€๋ฅผ ๋‹ค์‹œ 304๊ฐœ์”ฉ 12๊ฐ€์ง€๋กœ ๋‚˜๋ˆ„์–ด 6๋ช…์˜ ์‚ฌ๋žŒ์—๊ฒŒ ์‹คํ—˜ํ–ˆ๋‹ค.

 

Clutter๋ฅผ ์ธก์ •ํ•˜๋Š” 3๊ฐ€์ง€ ๋ฐฉ์‹์— ๋”ฐ๋ผ ์ธก์ •ํ•œ ๊ฐ๊ฐ์˜ clutter์™€ Log(RT) ๊ฐ’์„ ๋น„๊ตํ–ˆ๋‹ค.

๊ฒฐ๊ณผ์ ์œผ๋กœ, ๋ชจ๋“  ์ธก์ •๋ฒ•์— ๋Œ€ํ•ด clutter ๊ฐ’์ด ํด์ˆ˜๋ก ๋ชฉํ‘œ๋ฅผ ์ฐพ๋Š” ๋ฐ์— ๋” ๋งŽ์€ ์‹œ๊ฐ„์ด ๊ฑธ๋ ธ๋‹ค.

 

 

 19๊ฐ€์ง€ ์ด๋ฏธ์ง€๋งˆ๋‹ค target์˜ ์ข…๋ฅ˜์™€ location์— ๋”ฐ๋ฅธ Reaction time์˜ ํ‰๊ท ์„ ๊ตฌํ•ด, target absent/present ๋งˆ๋‹ค clutter์— ๋”ฐ๋ฅธ Reaction time์˜ ๋กœ๊ทธ ๊ฐ’์„ ๋ณด์—ฌ์ฃผ๊ณ  ์žˆ๋‹ค.

 Feature congestion์˜ ๊ฒฝ์šฐ correlation coefficient๊ฐ€ target present์ผ ๋•Œ .74, absent์ผ ๋•Œ .76์ด๋‹ค. Subband Entropy์˜ ๊ฒฝ์šฐ target present์ผ ๋•Œ .75, absent์ผ ๋•Œ .75์ด๋‹ค. Edge density์˜ ๊ฒฝ์šฐ target present์ผ ๋•Œ, absent์ผ ๋•Œ ๋ชจ๋‘ .83์ด๋‹ค. ๊ฒฐ๊ตญ ๋ชจ๋“  ์ธก์ •๋ฒ•์— ๋Œ€ํ•ด ์œ ์˜๋ฏธํ•œ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

 

๋”ฐ๋ผ์„œ clutter๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ reaction time์ด ์ฆ๊ฐ€ํ•˜๋ฏ€๋กœ,

clutter์— ๋”ฐ๋ฅธ reaction time์„ ์ธก์ •ํ•˜๋Š” ์ƒํ™ฉ์—์„œ๋Š” ๊ฐ ์ธก์ •๋ฒ•์ด ๋ชจ๋‘ ์ ์ ˆํ•˜๋‹ค๊ณ  ํ‰๊ฐ€ํ–ˆ๋‹ค.

 

 

 

"Experiment 2: Contrast thresholds for visual search in maps"์—์„œ๋Š” ๊ฐ ์ฐธ๊ฐ€์ž๊ฐ€ ๋ชฉํ‘œ๋ฅผ ์‹๋ณ„ํ•  ์ˆ˜ ์žˆ๋Š” ์ตœ์†Œํ•œ์˜ ๋Œ€๋น„ ์ˆ˜์ค€์„ ์ฐพ๊ธฐ ์œ„ํ•ด target์˜ ๋Œ€๋น„๋ฅผ ์ ์ง„์ ์œผ๋กœ ์กฐ์ ˆํ•˜๋ฉด์„œ ์‹คํ—˜์„ ์ˆ˜ํ–‰ํ–ˆ๋‹ค.

 

 ์ด๋ฏธ์ง€๋Š” ํ™”๋ฉด์— 1์ดˆ๋™์•ˆ๋งŒ ๋ณด์ด๊ฒŒ ํ•˜๊ณ , ํ™”๋ฉด์— ๋‚˜ํƒ€๋‚˜๋Š” ํ™”์‚ดํ‘œ๊ฐ€ ์™ผ์ชฝ์„ ๊ฐ€๋ฆฌํ‚ค๋Š” ๊ฒƒ์ด์—ˆ๋Š”์ง€, ์˜ค๋ฅธ์ชฝ์„ ๊ฐ€๋ฆฌํ‚ค๋Š” ๊ฒƒ์ด์—ˆ๋Š”์ง€์— ๋”ฐ๋ผ ๊ฐ๊ฐ “d”, “k” ํ‚ค๋ฅผ ๋ˆ„๋ฅด๊ฒŒ ํ–ˆ๋‹ค. 20๊ฐ€์ง€ clutter์„ ๊ฐ–๋Š” ์ด๋ฏธ์ง€ ๋ณ„๋กœ target์˜ ์ข…๋ฅ˜์™€ location์— ๋”ฐ๋ฅธ contrast ์ž„๊ณ„์ ์˜ ํ‰๊ท ์„ ๊ตฌํ•ด ๊ทธ๋ž˜ํ”„๋กœ ๋‚˜ํƒ€๋ƒˆ๋‹ค. Clutter ์ธก์ •๋ฒ•์— ๋”ฐ๋ฅธ ๊ทธ๋ž˜ํ”„๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

 

 ๊ฒฐ๊ณผ์ ์œผ๋กœ, clutter๊ฐ€ ์ฆ๊ฐ€ํ•˜๋ฉด search๋ฅผ ์œ„ํ•ด ํ•„์š”ํ•œ contrast ์ž„๊ณ„ ๊ฐ’์ด ์ฆ๊ฐ€ํ–ˆ๋‹ค. Clutter ์ธก์ • ๋ฐฉ๋ฒ•์— ๋”ฐ๋ผ Feature congestion > Edge density > Subband entropy ์ˆœ์œผ๋กœ ์ƒ๊ด€ ๊ด€๊ณ„๊ฐ€ ๋” ์ž˜ ๋ˆˆ์— ๋„์—ˆ๋‹ค.

 

 

 

"Experiment 3: whether color variability matter?” ์—์„œ๋Š” color variability๊ฐ€ clutter ์ธก์ •์— ์–ด๋–ค ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š”์ง€ ์‹คํ—˜ํ•˜์˜€๋‹ค. ์šฐ์„  Edge density๋Š” ๋ช…๋„๋ฅผ ๊ธฐ์ค€์œผ๋กœ edge๋ฅผ ๊ตฌ๋ถ„ํ•˜๋ฏ€๋กœ color variability์™€๋Š” ๋ฌด๊ด€ํ•œ ์ธก์ •๋ฒ•์ด๋‹ค.

 

 Photoshop์„ ์ด์šฉํ•ด Map images๋ฅผ ๊ฐ๊ฐ gray map, red map์œผ๋กœ ๋งŒ๋“ค์–ด original๊ณผ์˜ clutter๋ฅผ๋น„๊ตํ–ˆ๋‹ค. ์‹คํ—˜์€ ์‹คํ—˜1๊ณผ ๋™์ผํ•˜๊ฒŒ Reaction time์„ ์ธก์ •ํ–ˆ๋‹ค. ์ด๋•Œ target๋งŒ green, yellow orange, gray ์„ธ ์ข…๋ฅ˜์˜ gabor๋กœ ๋ฐ”๋€Œ์—ˆ๋‹ค.

 Feature Congestion์œผ๋กœ ์ธก์ •ํ•˜๋ฉด red map, gray map ๋ณด๋‹ค original์ด ๋” clutteredํ•˜๊ณ , Subband Entropy๋กœ ์ธก์ •ํ•˜๋ฉด red > origin > gray ์ˆœ์œผ๋กœ clutteredํ–ˆ๋‹ค. ๋”ฐ๋ผ์„œ Feature Congestion ์ธก์ •์— ๊ธฐ๋ฐ˜ํ•œ ๊ฒฐ๊ณผ๊ฐ€ ๋‚˜์˜ค๋ ค๋ฉด red, gray๋ณด๋‹ค original์—์„œ reaction time์ด ๋” ์˜ค๋ž˜ ๊ฑธ๋ฆด ๊ฒƒ์ด๊ณ , Subband Entropy ์ธก์ •์— ๊ธฐ๋ฐ˜ํ•œ ๊ฒฐ๊ณผ๊ฐ€ ๋‚˜์˜ค๋ ค๋ฉด red map์—์„œ reaction time์ด ๋” ์˜ค๋ž˜ ๊ฑธ๋ฆด ๊ฒƒ์ด๋ผ๊ณ  predictionํ•˜์˜€๋‹ค.

 

 ์‹ค์ œ ์‹คํ—˜ ๊ฒฐ๊ณผ, target์ด ์žˆ๋“  ์—†๋“  red map ๋ณด๋‹ค original map์—์„œ reaction time์ด ์ปธ๋‹ค.

 ์ด๋ฅผ ํ†ตํ•ด color variability๋ฅผ ์ œํ•œํ•œ ์ƒํƒœ์—์„œ clutter์„ ์ธก์ •ํ•  ๋•Œ Feature Congestion ์ธก์ •๋ฒ•์€ ์ ์ ˆํ•˜์ง€๋งŒ, Subband Entropy๋กœ ์ธก์ •ํ•˜๋ฉด ์ ์ ˆํ•˜์ง€ ์•Š์€ ๊ฐ’์ด ๋‚˜์˜ฌ ๊ฒƒ์ž„์„ ์•Œ ์ˆ˜ ์žˆ์—ˆ๋‹ค.

 

 ๋˜ํ•œ Target์ด absent์ผ ๋•Œ์— ๋Œ€ํ•ด origin๊ณผ gray map ๊ฐ„์˜ reaction time์— ์ฐจ์ด๊ฐ€ ๊ฑฐ์˜ ์—†์—ˆ๋Š”๋ฐ, ๊ทธ ์ด์œ ๋Š” target ์ค‘ ํ•˜๋‚˜๊ฐ€ gray์—ฌ์„œ ๋ฐฐ๊ฒฝ์ด gray์ผ ๋•Œ target์ด ์—†์Œ์„ ํ™•์‹ ํ•˜๊ธฐ๊นŒ์ง€ ์‹œ๊ฐ„์ด ์˜ค๋ž˜ ๊ฑธ๋ ธ๊ธฐ ๋•Œ๋ฌธ์ผ ๊ฒƒ์œผ๋กœ ์ถ”๋ก ํ–ˆ๋‹ค. (์ด๋ฅผ ํ†ตํ•ด background map์˜ ์ƒ‰์ƒ๊ณผ target์˜ ์ƒ‰์ƒ์ด ์–ด๋–ป๊ฒŒ ๋‹ค๋ฅธ ์ง€๋„ search performance์— ์˜ํ–ฅ์„ ๋ฏธ์ณค์„ ๊ฒƒ์ž„์„ ์•Œ ์ˆ˜ ์žˆ๋Š”๋ฐ, ์ด ๋ถ€๋ถ„์€ ์‹คํ—˜ ์กฐ๊ฑด์„ ์–ด๋–ป๊ฒŒ ์„ค์ •ํ•˜๋Š”์ง€์— ๋”ฐ๋ผ ์ถฉ๋ถ„ํžˆ ๋‹ฌ๋ผ์งˆ ์ˆ˜ ์žˆ๋Š” ๊ฐ’์ด๋‹ค. ์ด์— ๋”ฐ๋ฅธ ํ›„ํ–‰ ์—ฐ๊ตฌ๊ฐ€ ๋” ํ•„์š”ํ•˜๋‹ค๊ณ  ๋Š๊ปด์กŒ๋‹ค.)

 

 


 

Conclusion and future work

 

 ์ด ์—ฐ๊ตฌ์—์„œ๋Š” ์„ธ ๊ฐ€์ง€ clutter ์ธก์ •๋ฒ•์„ ์ œ์•ˆํ•˜๊ณ  ์ด ์ธก์ •๋ฒ•๋“ค์ด ์—ฌ๋Ÿฌ ์ƒํ™ฉ์—์„œ ์ ์ ˆํ•œ ์ธก์ •๋ฒ•์œผ๋กœ ์“ฐ์ผ ์ˆ˜ ์žˆ๋Š”์ง€ ์‹คํ—˜ํ•˜์˜€๋‹ค. ์•ž์„œ ์–ธ๊ธ‰๋œ 3๊ฐœ์˜ experiment๋“ค ์™ธ์—๋„ ๋‹ค๋ฅธ ์—ฌ๋Ÿฌ ์ƒํ™ฉ๋“ค์—์„œ "Feature Congestion", "Subband Entropy", "Edge Density"๊ฐ€ ์œ ์šฉํ•œ์ง€ ์™ผ์ชฝ์˜ ํ‘œ๋กœ ๋‚˜ํƒ€๋ƒˆ๋‹ค. ๊ฒฐ๊ณผ์ ์œผ๋กœ ๋‹ค์–‘ํ•œ ์ƒํ™ฉ์—์„œ ์œ ์šฉํ•˜๊ฒŒ ์“ฐ์ผ ์ˆ˜ ์žˆ๋Š” ์ธก์ •๋ฒ•์€ “Feature Congestion”์ด๋‹ค.

 

 


 

๋žฉ์‹ค์—์„œ 7์›” ํ•œ๋‹ฌ ๊ฐ„ ๊ฒŒ์ž„ ํ™”๋ฉด์˜ visual clutter๋ฅผ ์ธก์ •ํ•˜๋Š” ์ž‘์—…์„ ์ง„ํ–‰ํ–ˆ๋Š”๋ฐ,

๊ทธ๋•Œ ์œ„ ๋…ผ๋ฌธ์—์„œ ์†Œ๊ฐœ๋œ ์ธก์ •๋ฐฉ์‹ ์ค‘ ํ•˜๋‚˜์ธ 'Feature congestion' measuring์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.

 

๋ฐฐํฌ๋œ ๋งคํŠธ๋žฉ ์ฝ”๋“œ๋ฅผ ์ด์šฉํ•ด fps ๊ฒŒ์ž„์—์„œ graphics์„ ์–ด๋–ป๊ฒŒ ์„ค์ •ํ•˜๋Š๋ƒ์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง€๋Š” visual clutter๋ฅผ ์ธก์ •ํ–ˆ์Šต๋‹ˆ๋‹ค.

์ด์— ๋”ฐ๋ฅธ ๋ถ„์„ ๊ฒฐ๊ณผ๋Š” ๋‹ค์Œ ํฌ์ŠคํŠธ์— ์ž‘์„ฑํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค!