Meeting Info

Notes from the meeting on June 5th, 2008

Video:
- 30 short videos, where 10 are outdoors, 10 indoors, and 10 animation
The outdoors shots will use 1 camera only.
The indoor shots will use 2 cameras. The overhead angle between the two cameras must be less than 90 degrees, to allow for an overlap between the focused object. This will allow us to obtain 3d data concerning the focused object. Additionally, we should use the same camera for data output similarity.
The animated videos will use 3 cameras.
All of the videos will contain some element that is in motion. They will vary in length from 5-10 seconds.
We will consider different types of lighting as well.

Currently, we have decided to record to miniDV. The output should be in avi or Quicktime format. From my understanding, I believe that converting to Flash will be easiest to display on a webpage, and will be of the smallest file type for users to handle.

Survey:
The options for the survey have been reduced down to 2 surveys.
Survey 1 : Ask the user if the images are similar, and then ask on a scale of 1-5 of how similar (complete)
Survey 2: Present 6 images to the user, ask for two similar images, and then ask questions. (complete)

For survey 2, the questions will be based on shape, color, subject, lightness/darkness, motion and possibly texture and emotion.

Updates

The website now is currently being redesigned. I'm learning CSS in order to create a more visually appealing and interesting site.

Here is a preview so far:

Updates

- An index page has been added that links to all three surveys.

Image survery no. 2:
- If "yes" is selected for similarity, pressing submit directs the user to another page. This page displays the images again, and currently asks the user to rate the similarity on a scale of 0-3.
- If "no" is selected, pressing submit goes to the completed page.
- In both cases, the data is recorded in the table.

Fixes and more pages

The database is working properly now. All variables are being transferred to the processform page, where both the image tags and user input are stored in a single row. This fixes the problem created by the old version, in which refreshing the page would submit the image tags, regardless of whether the user submitted the form.

Current Pages:
1. Display of two random images, asks visitor multiple structural questions
2. Display of two random images, asks visitor (yes/no) if the images are similar
3. Display of six random images, asks visitor to find the two most similar

Updates:
- The layout of 1 and 2 have been changed to display both images in a single row.
- All images link to their original size

Soon to come:
- page with one constant image, and several random images. will ask user to select a similar image to the constant, and if user does not find one, the user can grab a new batch of random images.
- video comparisons

Image comparisons

Elements of art:
- Line - visual path taken by the eye within the piece
- Shape - areas defined by edges within the piece
- Value - lightness and darkness (tint, tone, shade)
- Texture - surface qualities
- Color - hues, various values and intensities
- Direction - visual routes taking vertical, horizontal or diagonal paths
- Perspective - expression of depth

Composition consists of :
- shape and proportion
- balance
- harmony among elements
- orientation of elements
- cropping
- negative space
- color
- contrast
- geometry
- rhythm
- lighting, illumination
- repetition
- perspective
- tension

As of now the site questions visitors on shape, color, tone, composition, and pattern. The inputs are yes and no (binary) answers.

Collecting data

The overall purpose of this portion of the project is to collect data from visitors. The data collected will tell us which attributes cause an image to seem similar or analogous to other images. The goal of this project is to collect enough data in order to judge the importance of certain attributes over another. These attributes will give a general idea on what type of image analysis software we must build. With various ways to analyze an image, it will be possible to build a search engine that groups certain types of images together based on their overall similarity or analogousness.

Current features include:
- Retrieving 2 random image url from database (imgid)
- Web page that displays the 2 random images
- Questions visitor on a scale of 0-3 how similar the images are
- Stores in 2 seperate database files (uimg, feedback) the imgid values, and the submitted 0-3 value
- Submit page allows user to return to original page