Well, I'd like to submit this PDF for your approval
What do you think?
Firstly, I don't ever want to sound like I'm criticizing the efforts of people who put a lot of work into scanning old books to preserve them. Especially not forum members. The direction I'm coming from is that the available technology has a lot of weaknesses, and I'm trying to build a reference manual to help people get the best result for their effort. Especially since many well-meaning people are not experts in all the relevant stages of the workflow to produce good quality electronic documents (and neither am I.) Not to mention that some of the tools needed are not freely available. And some of the flaws I criticize in many works I see are things I don't yet have a solution workflow/toolset for myself. (eg mixed screened image and sharp text/line overlays.)
I also don't want to get into the business of being asked to judge stuff like this. But since you asked... some thoughts as I read through the PDF. (I've temporarily misplaced my djvu reader, doh. So can't see that file just now.)
It's a PDF. Won't go into the merits/flaws of PDF here, and it's great you provided an alternative.
The cover. Nice and clean. You get the concept of removing superfluous background detail since it bloats file size. But... there are a few places where 'visual reality' has historical value, and the cover is one of them. You've routinely OCR'd all text, cover included. Result is the cover titles look too sharp and there's no sense of how they really looked. I'd have reduced the background to something like yours, but kept the title text as true image. Also kept some or all of the comb binder. So the cover at least still looks like the actual book. In 100 years, no one alive will remember what these really looked like.
When some particular element has detail you consider important (as you recognize for the cover texture) it's worth allocating some more filesize to it. Because you substituted a background, the Tek silver logo now looks like a sharp-edged insert (because it is.) Might have been better to do the cover as a high-res scan, then scale the whole thing to the wanted size. Up to you whether that includes the cover texture dots, but the scaling will keep the text and logo edges looking real.
In general I'm wary of fully OCR-converted text, since it's notoriously error-prone, and there's no fallback to the original image. One can't trust it as a historical reference. But there isn't any good format I'm aware of that bundles both the original page image plus the OCR text, with full word by word linkage.
First few pages with grayscale photos and text. Nice job removing the screening pattern without moire.
The Warranty and Contents images have a different shade balance to the pic of the unit though. I don't have that manual so don't know how it is really, but I think maybe you used different processing on them. It's important to experiment, find a processing recipe that works for all the images in a book (or sets of similar), then stick to that exactly during the workflow. To avoid this kind of uncertainty of whether the images were or weren't actually different looking.
That the main Contents page is a static image, that doesn't work as real clickable items to get to book sections, is a flaw with the PDF tools. Other methods can do it, like html image hot areas, etc. So can PDF with authoring tools. Yeah I know, after spending many days scanning and cleaning, making a working index is a pain. But its lack is also going to be a pain for everyone using the ebook, forever.
The main text OCR is very nicely done. Superscripts, fonts, etc correct. So nice I suspect you did a lot of manual cleanup? Since the text is totally abstracted, there's no paper and ink blemishes at all. Skimming, didn't spot any obvious OCR-errors. Even the equations are perfect. Did you manually proofread or hand-layout the entire thing? Vastly better than most OCR efforts.
Included 'Notes' pages. Yay! Visual veracity ftw.
Page 2.2 pic of the unit is again different tonal quality to the earlier one. Also there is some bleedthrough of paper other side text. Hint: to avoid that, use a matt black backing sheet when scanning. Black felt works well.
Page 2-5, 6-2, etc line drawings are half-toned, ie fax mode. The edge jagginess even at normal zoom is noticeable. In zoom it's nasty. Better way to do these is scan in 256 level gray, then adjust color curve to get saturated black and white, but preserve gray scale on edges. Clean any remaining image blemishes, then scale to final size, then reduce number of PNG bits/pixel to minimum that preserves visual line edge shading to look clean. With a clean white background, coding as PNG with these steps achieves surprisingly small filesize.
This goes for all the subsequent B&W line drawing inserts. Worst case: waveforms on 2.22 _really_ missing gray scale.
Page 2-15 table. Yeow, you did that by hand in MS Word? Lots of work I expect, but... the vertical alignment of text vs the bars is broken. So is text v-alignment in boxes. Not your fault, I know it's the tools. Which is the kind of thing that pisses me off - a utility to do this right should start with the base scanned image, and allow OCR and auto-aligning of the abstracted text with the original text elements. Sigh. I'm sure it was correct in the paper book. So this kind of visually awful result makes the original layout artists and typesetters look bad.
Page 4-2, 4-7, etc. The dreaded problem: text and line drawing combined with screened shading. You can't blur the shading adequately to get rid of the screening pattern and posterize without moire, while avoiding ruining the crisp line edges of the illustration & text. I'm still looking for a descreening utility that deals with this properly. Got one that's _almost_ there. It uses Fourier analysis to identify the screened areas. But then insists on doing the descreen itself, instead of just passing a mask to photoshop to allow manual tweeking as required. Needs more work.
Anyway, your image has background moire patterning (wasting file size) but it's not your fault.
All the PCB/guts photos eg pg 4-4 - pretty good, nice tonal range and clean. I'd have adjusted for clean white backgrounds and PNG-coded to avoid JPG artefacts (zoom on blank areas to see what I mean.) Also for PNG-grayscale you can choose the number of bits per pixel, and much less than 8 is often perfectly adequate. Tradeoff between visible posterization of shaded areas, and filesize. PNG doesn't suffer from the artefacting JPG does, but being non-lossy, decent filesize is dependent on your pre-processing to remove superfluous noise in blank image areas.
Block diagram - nice, perfectly aligned. Same comments about fax mode vs keeping some gray on edges.
'Voltage and waveform test conditions' - in blue. Cool! But I'll be super picky: it's a different blue to that on the schematics, and I'm sure in the book it's the same ink.
Schematics: Yay for generous resolution to keep all annotations easily readable.
Same comment about 2-tone jaggies. Interesting though - the blue color means it actually has 3 colors - wht, blk, blu. Pity the pic is PDF-embedded. I'm curious to see how it's coded, and later I'll extract it with photoshop to have a look. Not that I'll find out how it's coded internally in the pdf.
Many of the schematic sheets are quite skew - did you stitch and still forget to deskew? Or was that missing in your tools? What were you using to stitch? I can't see the join, so it worked well. But "small window"? What?
Finally: This wall of text doesn't mean I think your Tek1S1 etext was bad. Your PDF is actually one of the best I've seen. I'm just a picky perfectionist and think these old manuals deserve a much better preservation in perpetuity than our current software tools and document formats allow.
Also I suppose that my mum being a commercial artist, and my assisting her in the past with software book layout issues, contributed to my dissatisfaction factor with most technical etexts.
Home
Help
Search
Login
Register
