WebP near-lossless

[Vikas Arora]

Hi -

We have recently added a new compression option (-near_lossless) to the WebP encoder (cwebp) that will pre-process the input image by adjusting the pixel values and produce compacter images. The output bit-stream is compatible with the WebP-lossless mode (no need to specify '-lossless' option with cwebp) and can be decoded with the standard WebP (lossless) decoder.

This new option (cwebp -near_lossless) is specified with an integer parameter with a range 0 (maximum pre-processing) to 100 (no pre-processing). 

Following is the typical rate-distortion matrix for this new compression option:

-near_lossless 100: No-loss (bit-stream same as -lossless).

-near_lossless  80: Very very high PSNR (around 54dB) and gets an additional 5-10% size reduction over WebP-lossless image.

-near_lossless  60: Very high PSNR (around 48dB) and gets an additional 20%-25% size reduction over WebP-lossless image.

-near_lossless  40: High PSNR (around 42dB) and gets an additional 30-35% size reduction over WebP-lossless image.

-near_lossless  20 (and below): Moderate PSNR (around 36dB) and gets an additional 40-50% size reduction over WebP-lossless image.

Attaching Google Logo images (Google_logo.n.webp) compressed with various -near_lossless values.

This new compression mode is found useful for compressing graphical images. Please try this new compression option on your favorite graphical images and see if this is useful in saving image bytes further without impacting visual quality for those images.

Thanks,

Vikas