Sorting and Searching Behavior Patterns: Analyzing WHR Arrays with the Selection Sort Algorithm ISHE 2010 Video Enhanced poster

Abstract:

The goal of this project is to develop a ‘peak-shift’ index to assess the attractiveness of behavior patterns, rather than images alone. Waist-to-hip ratio arrays comprised of data collected from video recordings of behavior patterns (e.g., walking, dancing) will be sorted using the selection sort algorithm. These view-dependent WHRs, derived from unknown actual-WHRs, will be used to create a ‘peak-shift’ index. Sorted arrays will be used to find image data of: 1) less than .69, the peak-shifted elements, 2) .70-.72 intermediate elements and 3) greater than .72 elements. In short, data collected from behavior patterns will be sorted and searched. The hypothesis that lower WHR elements increase viewing time will be tested.

Location: Poster session at the International Society for Human Ethology Conference, Madison, WI, 2010 (poster)

Materials and Methods: Behavior pattern analysis

1) Background - Measuring WHRs from Behavior Patterns: Actual vs. View-Dependent WHR. Actual waist-to-hip ratios are calculated using the circumference of the waist divided by the circumference of the hips. Though this information may be a general description of body shape and useful for evaluating the attractiveness of still images, it has been argued that actual-WHR is not information that is available to viewers and that WHRs change as a function of viewing perspective and the movement of the person viewed. View-dependent WHRs are derived from measurements taken across the narrowest section of the waist and widest section of the hips. These view-dependent WHRs more accurately represent the WHRs from behavior patterns that individuals actually see (Doyle 2009a, b) and have been shown to be rated as differentially attractive in a small study (Doyle 2009c).

Waist to Hip Ratio Measures:

Actual-WHR = a measure of the circumference of the waist/hips.

View-Dependent WHR Measures:

  • View-dependent measures are taken from anatomical markers, such as the waist and hips, in frontal, profile, side and posterior orientations.
  • Across-WHR = is a view-dependent length measure taken across the waist/hips. Exp: frontal-view WHR is a measure across the body and is view-dependent.
  • Left & Right-WHR = length across the waist and hips measured from a medially placed, vertical centerline.

 

 

2) Measuring view-dependent waist-to-hip ratio (vdWHR) in behavior. 

Example screen captures of the five sampled videos. Five videos were sampled at approx. 30 frames/second. These frame samples (i.e., the images) were measured to derive real-time view-dependent WHRs. The sampled images were then used to create looping videos between 1-2 seconds in duration that the study participants viewed. 

Walking Video 1 (Posterior-View)

Walking Video 1 (Posterior-View)

Belly Dancing Video 1  (Fast)

Belly Dancing Video 1
(Fast)

Belly Dancing Video 2 (Slow)

Belly Dancing Video 2 (Slow)

Belly Dancing Video 3  (Profile)

Belly Dancing Video 3
(Profile)

Walking Video 2 (Frontal-View)

Walking Video 2 (Frontal-View)

 

 

 

 

Example of sampled images - Video 1, Walking (Back-view).

Video 1, Walking (Back-View) was composed of 31 sampled and measured images.

Video 1, Walking (Back-View) was composed of 31 sampled and measured images.

 
 

View-dependent waist-to-hip ratio (vdWHR) in behavior: Across-, Left- and Right- side vdWHRs. To isolate components of WHRs in motion the images were also divided along medially-placed vertical centerlines to create left- and right-side view-dependent WHR versions of each video.

The left and right-side videos derived from Video 1, Walking (Back-View) were each composed of 31 sampled and measured images. 

Example of Left-Side view from Video 1, Walking (Back-View)

Example of Left-Side view from Video 1, Walking (Back-View)


  
 

Example screen shots from the left- and right- side vdWHRs videos.

back_Left back_Right BD_Left BD_Right BD2_Left BD2_Right BD_ProfileLeft BD_Profile_Right front_Left front_Right

3) The real-time vdWHRs were then plotted on graphs. The graphs show in real-time the vdWHR in the walking and dancing videos. Notice the alternating high and supernormally-low vdWHR shown to be present. It was previously hypothesized that such stimulus presentation from behavior patterns (e.g., walking and dancing) could create peak shift effects (Doyle 2009a, b, c.).   

Example graph of all View-Dependent WHRs from Video 1, Walking (Back-View) Behavior Pattern.

The across, left and right- WHRs are depicted as they occur within the behavior pattern. The highest, (least attractive), and lowest (most attractive) view dependent WHRs are labeled. Alternating sinusoidal sides are clearly visible (Doyle 2009a, b, c).

The across, left and right- WHRs are depicted as they occur within the behavior pattern. The highest, (least attractive), and lowest (most attractive) view dependent WHRs are labeled. Alternating sinusoidal sides are clearly visible (Doyle 2009a, b, c).

 

4) Selection Sort: Analogy – Behavior patterns contain information* which creates (WHR) stimuli arrays within the environment. The information from these behavior pattern arrays is analogous to information located within computers: it can be sequentially searched and sorted. Information from behavior patterns may also be searched and sorted.

 * WHRs are information, distance across waist and distance across hips are data.

Selection Sort Algorithm:

          Find smallest value, swap with first position

          Find next smallest value, swap with next position

          Repeat...

The real-time vdWHRs were then sorted with a selection sort algorithm. The first three (left side) columns show the unsorted real-time vdWHRs (across-, left- and right-side vdWHRs). The right side columns (4th, 5th and 6th) show the sorted vdWHRs (across-, left- and right-side vdWHRs, respectively). The columns are color-coded: 1) Red/orange - less than .70, the peak-shifted elements, 2) Salmon - .70-.72 intermediate elements and 3) Pink - greater than .72 elements. The colored WHR sorts show at a glance the proportion of super-low vdWHR stimuli being presented.

S++, S+, S-

Videos: Walking and Dancing Videos showing graphed vdWHRs in real-time. Watch the video of each behavior pattern next to the real-time vdWHRs corresponding to each selection sort by following the link next to the video. Or see the Playlist.

Video and real-time vdWHRs graphed, Walking (Back-View) Watchback_All 

 

Video and real-time vdWHRs graphed, Video 2, Belly Dancing (Front-View). Watchbelly_Dance_All

 

                                    
Video and real-time vdWHRs graphed, Video 3, Belly Dancing (Front-View). Watchbelly2_All 

 

Video and real-time vdWHRs graphed, Video 4, Belly Dancing (Profile-View) Watch.belly_Dance_Profile_All 

 

Video and real-time vdWHRs graphed, Video 5, Walking (Front-View). Watchfrontal_All   

5) Results So, are there significant view-dependent WHR differences within and between the behavior patterns?

Yes:

Example: Within Behavior Patterns mean WHRs, Across- vs. Left-, Across- vs. Right & Left- vs. Right

Within_Between_Paired_Samples

10 of the 15 Across, Left & Right WHRs are significantly different.

Example: Between Behavior Patterns Mean Across-WHRs

between_Behavior_Patterns

Only the walking back-view vs. walking front-View, the 2nd Belly Dancing video vs. the Profile Belly dancing videos did not significantly differ. Note that many >= .69 WHRs have significantly different means. *Between Behavior Pattern mean WHRs Across- vs. Left-, Across vs. Right & Left vs. Right omitted.

Between behavior patterns, all mean WHRs, all Across, all Left and all Right.

all across, left, right WHRs

So, what about participant viewing times, were they significantly different for the videos? Yes and No…

viewing_times

Within behavior patterns, mean viewing times, Across- vs. Left-, Across- vs. Right. *Only 4 of 15 pairs significantly different. *Within behavior patterns, mean viewing times, Left- vs. Right. Not significantly different. *Between behavior patterns, mean viewing times, across-WHRs. Not significantly different.

vtresults

 

 

 

Between Behavior Patterns Mean viewing times for All Across- vs. All  Left-, All Across- vs. All Right & All Left- vs. All Right WHRs.

Between Behavior Patterns Mean viewing times for All Across- vs. All Left-, All Across- vs. All Right & All Left- vs. All Right WHRs.

Between behavior patterns, mean viewing times, Left vs. Right-WHRs. Significant:

b_View_Times_L_R

Attractiveness Ratings (1-7)

Within Behavior Patterns

img75

Between Behavior Patterns

img77

Inconclusive Pilot Study. Viewing time did not seem to be systematically affected by view-dependent WHRs differences, even for those WHRs that were significantly different, as lower WHRs were viewed significantly longer. Attractiveness rating faired somewhat better. Though behavior patterns may be searched and sorted, it was not not possible to create a "Peak-Shift" index. Back to the drawing board...

References

Doyle, J. (2009a). A Woman's Walk: Attractiveness in Motion. The Journal of Social, Evolutionary, and Cultural Psychology, 3(2), 81-92. (Article)

Doyle, J. F. (2009b). A woman’s walk: Attractiveness in motion. Poster session presented  at the North Eastern Evolutionary Psychology Conference, Oswego, NY. (Abstract) (Poster)

Doyle, J. F. (2009c, unpublished manuscript). Physical Attractiveness Ranges: Standing and Contraposto Pose Waist-to-Hip Ratio in 360° and Pairs. (Manuscript)