Vision Processing Flashcards
Laboratory Material
What does SPV stand for and how has it been useful?
Synthetic Prosthetic Vision
- useful predictive tool in bionic vision research
- SPV presents a representation of bionic vision - what researchers believe someone might see with an implant
- allows for measurement of speed, accuracy, and other metrics
What are Phosphenes?
Visual percepts from a bionic eye are typically called phosphenes; individual pixels in code
Describe Regional Averaging
When an image has been broken into individual elements, and the average intensity of each element is considered to be the intensity of the entire phosphene.
A visual scene must be updated fast enough so that it doesn’t “flicker” and make the implant recipient motion sick.
This would mean the electrodes need to be updated at a rate of around __ times per second?
This would mean the electrodes need to be updated at a rate of around 50 times per second.
For 100 electrodes, that means 5000 instructions that need to be calculated, sent to the implant and delivered as a stimulus every second.
To address the speed issue, one approach to image processing could be to instead of using Regional Averaging, we could use what?
Describe the alternate technique.
Impulse Sampling
Similar to regional averaging, but instead of deciding the intensity based on the whole region, a representative sample is taken - typically the pixel intensity at the centre of the region - to determine the intensity at which to convey the phosphene.
Describe Gaussian Sampling filtering.
A combination of Regional Averaging and Impulse Sampling, greater weighting to the pixels at the centre of the section is assigned but the intensities of the other pixels in the section is also taken into consideration; a weighting scheme is applied based on how far offsent from the centre a pixel is located in each section.
This weighting is related to the gaussian function which incorporates both mean and standard deviation into value calculation.
Ideally, one would use impulse sampled intensities of each pixel in an image and have this value represented by its own electrode.
Why is this not achievable?
Unfortunately, conveying this many “phosphenes” to a person has not yet been achievable (far from it). To do so would require over 110,000 electrodes behaving independently, i.e. with no “cross-talk” between them because of overlapping electric fields.
First generation technologies are still in the realm of being able to present around only 100 phosphenes. This is why image processing needs to be applied: to provide the most useful information to the implant recipient.
What is an image kernel?
A small matrix used to apply effects like the ones you might find in Photoshop/Gimp, such as blurring, sharpening, outlining or embossing.
They’re also used in machine learning for “feature extraction”, a technique for determining the most important portions of an image. In this context, the process is referred to more generally as “convolution”.
