These are the aspects that I think would provide the most benefit for a general range of uses. Where I have made reference to actual functions, I'm referering to the Magickwand api for C, found here:
http://www.imagemagick.org/script/magick-wand.php
Format conversion (MagickReadImage/MagickWriteImage)
Resizing/scaling - (MagickScaleImage/MagickResizeImage). You can choose the specific resizing algorithm you want, and resizing bitmap images gives better results than dbp, but more importantly, the resize features enable you to resize vector images, which is impossible in dbp.
Cropping, flipping and rotating - these kinds of things are possible in DBPro using sprites, but you then have to use the get image command to output it as an image or manipulate it further. Using 'get image' reduces quality, (I think it applies filtering, which alters the raw data of the image). Using memblocks can get tricky - so these functions would be a useful bonus.
Vector and text drawing. Things like DrawLine, DrawEllipse, DrawBezier, DrawRectangle, DrawAnnotation. Cloggy's d3d plugin is good for some of these, but has some issues in my experience - it works for me, but I had to remove it entirely from GUI Studio because it caused errors for other users. The options for drawing and text in Imagemagick are very versatile, too, with control over stroke width, fill colour, gradient fills etc. I'd especially like the text features, which would also mean wrapping the font and allignment methods.
Image effects: MagickGaussianBlurImage would be very useful - much more control than dbpro's blur bitmap command. MagickShadowImage (applies a drop shadow), MagickShadeImage (applies a distant light source to the image) and MagickRaiseImage (applies a sort of bevel) would be very nice features too.
Image Layering: Using the two functions MagickCompositeImage and MagickSetImageOpacity enable layering of one image over another - and there are a number of useful ways images can be added, like multiply, screen, difference etc. Could be done with memblocks, but very tricky.
There would also need to be a number of functions included for general functionality - things like changing stroke colour, creating/destroying/rendering 'wands' (from what I understand, wands are like separate images which can be named, and combined together - there are three varieties - magick wands, for image effects, drawing wands for vector drawing, and pixel wands for things like changing ink colour.) Also, for the Magickwand api at least, you need the functions 'MagickWandGenesis' and 'MagickWandTerminus' to inialise and terminate the process.
I think Ruccus has a good point - if someone like Ian would be willing to at least start this off, then I think that's probably the most difficult part. Once the framework is in place, anyone with a suitable compiler should be able to add functions reasonably easily, if the project is left open.