View Single Post
Old 15 February 2017, 18:59   #48
StatMat
Code Fiddler
 
StatMat's Avatar
 
Join Date: Apr 2016
Location: UK
Age: 48
Posts: 81
Okay, so indeed the PhotoShop dithering and quantization was producing better results (surprise, surprise! ). Therefore I have modified my original approach to more closely match PhotoShop as follows:

1. I am now using Floyd-Steinberg dithering
2. I have introduced a 25% reduction in the dithering error propagation to more closely match the default Diffusion dither with an amount of 75% in PhotoShop
3. I am now using a selective palette for the dithering based on each image rather than using a general purpose palette

Whilst my results do not match PhotoShop exactly (this is always going to be difficult given the number of variables involved and closed nature of the product), I believe it's now producing significantly better results than before that are rather comparable to PhotoShop. For example, here are the Turrican II and Battle Squadron images from before and after alongside Gibs original PhotoShop images:

StatMat (Before):



StatMat (After):



Gibs (PhotoShop):



StatMat (Before):



StatMat (After):



Gibs (PhotoShop):



This time I have uploaded a new pack to the FTP containing just the boxart and a combined v0.3 complete pack. Again, this is still PAL only at the moment. Let me know what you all think.
StatMat is offline  
 
Page generated in 0.14102 seconds with 11 queries