Project Abstract

I chose this topic after discussion with my mentor, as I found salience maps interesting and made a link between them and image compression – save storage space by compressing the parts of the image that are less visually important more, and maintain quality by compressing the visually important parts less. The digital space is becoming more and more important, with screen resolutions ever on the rise. This sort of dynamic compression could help businesses and other online platforms use their limited server storage space more efficiently. Additionally, with mobile browsing becoming ever more important, a smaller file size, for an image of the same apparent quality will reduce loading times, especially for those people using a mobile connection. I used the DeepGaze IIE model to generate the relevant salience maps, and utilised these within a Python script that manually applies the JPEG compression algorithm, using the heatmap to determine compression across the image. I have discovered that in general the principle is sound, there are a few limitations in my process that prevent me from declaring complete success. However there are strong indications that the concept holds promise.

Keywords: Computer Vision, Salience Maps, Image Compression

 Conference Details

Session: Poster Session A at Poster Stand 111

Location: Sir Stanley Clarke Auditorium at Tuesday 7th 13:30 – 17:00

Markers: Joe MacInnes, Daniele Cafolla

Course: BSc Computer Science, 3rd Year

Future Plans: I’m looking for work