Friday 26 February 2016

Image Visualisation

Something I have found quite surprising over the last few months is just how varied archives can be. I was recently given a large set of images to work with, which is not so usual, but the task at hand was trying to look at the items in a new way. This was only really possible in this case because the dataset I was working with was a group of digital images, this gives a lot of freedom in regards to quickly arranging and presenting large quantities of data. Providing our users a different way of viewing data is in many ways is just like supplying another means of access, it allows them to see things differently and draw their own conclusions without imparting any personal bias or inferences on our part.

Tasked with finding free software or tool to use to visualise the material I set about some serious desktop research. Whilst the work is ongoing here are my initial thoughts using two different pieces of software.


ImageSorterV4 dataset arranged via colour
ImageSorter
The first, ImageSorterV4 by Pixolution, allows a very quick and easy way to sort and arrange large volumes of images. It is limited in the sorting options available since it can only arrange by colour, name date and size. Of the choices available colour and date ate the two that are the most useful for our purposes with the colour option being very visually striking. The advantage of this program is just how quick it is, within locating the image folder it will process and display. 

Dataset arranged by Brightness vs Saturation 
ImagePlot
The second program ImagePlot by the Software Studies Initiative builds upon ImageJ a java based open source image processor. ImagePlot is a far more versatile program allowing a comparison of any two elements in relation to the image, providing that you already have access to the information in advance. At its base it is simply a scatter plot program, comparing two different values but the way this program differs is that once it has actually plotted the position it will place the relevant image at that location. This doesn’t sound ground breaking but it turns a very static and uninspiring looking scatter graph into a very visual experience. But it is still just a scatter plot so that means that as long you have two [pieces of data worth comparing then you can visually display this, so the focus shifts from looking for the right program to display your data to needing to find the right way of pulling data from the images themselves.

Metadata
Since the only information that we have available in this instance is the image itself we have to look to see what metadata we can extract from the file itself including file size or the date it was created. Most digital cameras also record information at the moment when the shutter-release button is pressed. Information such as, time and date, camera model, focus settings, FOV (field of view), was a flash used and in some cameras it will even record the GPS location of where the photo was actually taken. 

For my testing purposes using the ImagePlot software I used some easily accessible metadata in the form of name, date last modified and size of the file these I was able to collect in moments by using an internet browser. With files in a folder you can just copy and paste the location address into a browser window in order to get access to a simplified file hierarchy. Google Chrome in particular works well with this.

Image Analysis
Dataset showing the count of objects vs the percentage area
covered
If we want to do some useful comparisons we are going to need some more interesting data. The content of the images hold the most potential for relevant data. We can already view the images, but pulling quantifiable statistics for the purpose of comparison is a little more difficult. Thankfully ImagePlot actually comes with three additional macros that allow the user to pull a limited amount of this information from the images themselves. In this case I used these macros to pull information about that brightness, saturation, hue, count (number of objects greater than 10px) and the percentage of area cover (by the same objects). The count macro actually allows for a certain amount of variation in the way it collects its values, by default it is setup to search for ‘objects’ that are 10 pixels or larger so some complex images like people’s faces or textured backgrounds can sometimes return a value in the thousands but this setting can be altered to a higher number (say 100px) and it would only detect larger objects at the cost of any small objects getting overlooked. There is also a setting to change whether it is looking for more circular or rectangular sections to determine what is classes as an ‘object’ for the purposes of counting.

File Format
Once all this information has been gathered it simply needs to be placed into an excel spreadsheet with each row relating to a particular image. All of the methods I have mentioned so far specifically allow the information to be retrieved in text form and placed into a spreadsheet very easily. Once the spreadsheet has all the data that you wish to compare it is then saved as a tab delineated text file (.txt) which is the format that ImagePlot uses for running its scatter plot algorithm This freedom of just saving data into an excel spreadsheet and then turning that into a text files really opens up what you can do with ImagePlot since as long as you can get information and values you have something to compare and you don’t have to deal with a third party file type you’ve never encountered before.

Conversion of excel spreadsheet to tab delineated text file
Conclusion
Programs such as these are only limited by the data that can be extracted from the images themselves, the fact that I can only compare two statistics at a time and the manner that the data is displayed will always be a direct comparison of one element to another. It is also far easier to compare information that is pure numbers as opposed to text for example using the file name or any string/boolean value, although that is only specifically with the software that I have found so far. If there is software that could group images by textual information I would love to hear about it. 

With what I have so far I have a significant amount of information in regards to displaying the data, there are some comparisons that are most relevant than others and some that are far more visually appealing but it is the fact that I can show this data at all that is interesting. Being able to get a visual representation of data makes it easier to consider it as a whole and allows other people to draw conclusions about the entire dataset in a way that would otherwise be almost impossible only viewing small sections of the data at a time.

David Heelas
Transforming Archives Trainee

No comments:

Post a Comment

Comments and feedback welcome!