Optipress - Machine Learning based image compression
The main challenge with JPEG compression is to find the most optimal compression factor for each different image. Different images compressed with the same compression parameters may result in different perceived quality. Moreover, the same image in different sizes may require different compression strategies to achieve maximum size gain without visible quality loss.
To overcome this, we have developed the Optipress JPEG compression algorithm.
- Optipress finds out the best JPG compression approach by analysing specific image features and current compression parameters.
- A Machine Learning model determines the best compression strategy for this image.
- Quality is then evaluated based on a simulation model of the Human Visual System to achieve optimum compression without perceptive quality deterioration.
Usage
| optipress=1 | most conservative setting. Image quality is prioritised |
| optipress=2 | balanced setting |
| optipress=3 | most aggressive setting. Optimised for smaller file size |
Images created by Optipress fully comply with baseline JPEG specifications and are compatible with all JPEG encoders.
| filter value | file size |
|---|---|
| q=90 | 623 KB |
| q=85 | 531 KB |
| optipress=2 | 513 KB |
| optipress=3 | 482 KB |
In this example, Optipress increases compression to achieve better compression rate, compared to standard JPEG compression approach.
| filter value | file size |
|---|---|
| q=90 | 346 KB |
| q=85 | 231 KB |
| optipress=2 | 418 KB |
| optipress=3 | 342 KB |
With this image, Optipress detects faster perceptive quality degradation and compresses with more conservative approach.