Abstract:
"
With the increase demand for storage space across all devices, data compression and relating
standards have been receiving the buzz over the last five years. With the advancement of
computing power together with the abundance availability of ideal datasets has increased the
interests in the application of image processing and deep learning tasks. While the average user
may be ready to accept the standard JPEG encoder, a diversified group of users expects a better
image compression standard that can preserve more data and preserve a higher colour accuracy
while maintaining the same human visual systems with higher space saving. The author of this
research has implemented a content aware enabled image compression that will better preserve
the image subjects most valuable data while compressing the content around it, with the use of
deep learning and convolutional networks. This content aware feature enables to draw a region
of interest map around the uncompressed image by including a complete set of features from
most classes and after taking the threshold over the sum of all activation functions. The results
are tested using the industry standards of MS-SSIM and PSNR values to prove higher visual
quality can be preserved while compressing using this approach."
