Hacker News Reader

Comment by londons_explore | Hacker News Reader

Is it time to admit that simply storing everything as an image is the future?

Sure, it isn't technically as neat, but it gives far more flexibility. It happens already in memes (where text is usually part of a gif)

Text and text encoding dates back from the days when every byte of storage space mattered. Now lossless high resolution images are the norm, and wouldn't constrain what symbols could be included. If you want to make a word be slightly bendy for style, you can!

Accessibility, relayout, etc. can all be done with images too - you simply have a tool which extracts the text, transforms it however you please, and then puts it back as an image.

View on HN

I strongly disagree.

- You can copy a text symbol and put it into a search engine to figure out what it means.

- Screen readers can recognize text symbols without needing an alt text.

- Text is more flexible on dark or light backgrounds and has better contrast.

- Text is better suited for virtual keyboards and auto-correct or auto-replacement.

- Text can be used in input fields.

- Text has native rendering hints built-in, such as baseline and kerning, which improves inline rendering.

- Text is more flexible when choosing a glyph. For example, you can render it larger for the visually impaired, or change fonts and put a special glyph into the font.

- Text is more flexible when choosing color vs. black-and-white output. Try printing an emoji to see what I mean.

Tooling could be built to do all of that for images too.

For example you could take part of an image and put it into a search engine too... That search engine would take images as input, and where the image represents writing if some kind, it would return relevant results.

Screen readers could look into an image and read out any writing found, just like some screen readers can describe an image ("person in boat holding flag"). Deep nets that can do this reasonably well have been around years now.

The tooling for doing all that with images is called a font.

You can do all that but you'd have to use a neural net instead of "find..." Neural nets make plenty of errors, including on OCR (at least 2-4%).

Find my girlfriend a dress that doesn't need ironing.

Easy if there's a "do not iron" symbol in text to search. Much harder to grep if it's an image.

Fonts are a standard and cross-platform way to bundle vector images.

Now if we only had some way to figure out which image in the bundle corresponds to what... Maybe some encoding scheme where we assign a numeric ID to each image?

(Oh wait...)

you simply have a tool which extracts the text, transforms it however you please, and then puts it back as an image.

This is extremely non-simple. It is a strictly harder problem than just... encoding image+text in a sane container format like HTML or SVG.

But with various AI/ML approaches it's now becoming pretty straightforward. And lots of new hardware has ML acceleration.

So we train an ML model to convert images to text. How do we encode that text information?

This is the most absurd thing I've heard this month. The page you are looking at right now is in fact an image that your computer put on screen.

However, that fact does not help you when you want to reply to my comment. Or more generally, if you want to edit, transform, store, compress, search.

I can easily search 1GB of text on my computer for a specific pattern. I could not even store all that information directly encoded as (say) raster or vector graphics.

you simply have a tool which extracts the text, transforms it however you please, and then puts it back as an image.

Then you still need some way to handle text, and image to text is not reliable. Text simply has so much more distilled information. Images are nice for humans, but I can't imagine them as a storage format for programs.

How would the software that lets me add text to images work? Presumably it would use unicode.

And how would the software that extracts text from an image store the extracted text? If it comes accross a number of symbols from different domains how would it save them, and what table of contents would it use as a reference? Unicode.

But that's an implementation detail - some software might let you hand write text into an image with a brush or pen, and that might use as its internal representation the X,Y coordinates of the brush strokes.

So then how will the software convert the X,Y coordinates into text that is read out to a screen reader? I think many of the other comments put it more eloquently than I can.

Reply on news.ycombinator.com