Jim Williams, Circuits as art

I came across this great article about Jim William’s artistry as I searched through my EDN back issues. It is from the February 1987 edition of EDN.

I want to tell you all my scanning secrets in honor of my buddies Jim Williams and Bob Pease, may they both rest in peace. I put my scanning procedure after the article below. Bob and Jim never kept secrets about their work and I don’t do it with mine. If you don’t write articles or send Design Ideas into EDN because keeping them secret makes you think you indispensable at your job, remember what Charles de Gaulle said: “The cemeteries are full of indispensable men.” So, in honor of Jim, here is a great article about the art of analog design, more of which you can see at his landing page at the Computer History Museum. His lab bench is on exhibit there at least until April 15 2012. Click on any page image to get the 1024px image you can read.

.

And a free ad for Frequency Devices to boot. Remember their address and phone numbers may have changed since 1987.

I scanned it in at 600DPI in case I ever have time to OCR (optical character recognition) it. That is the first tip; always scan old EDN issues at 600DPI if you don’t want dozens of errors in a subsequent OCR. Lately I have been using the free PaperPort software that came with my Brother MFC-J6510DW printer, and it even works half decent at 300DPI. . It has an 11×17 format but I didn’t use it for this since darned if I can get it to stop helping me by color correcting, I just want the page. Also, like every Japanese scanner, the engineers at Brother have taken the pledge “I will never design a scanning product that scans the entire page, I will always cut of an eighth of an inch on all sides, this I pledge to my company, the Emperor and my ancestors.” So I used a much better scanner, my Canon CanoScan 8800D. I just run it through the Twain interface in the fantastic free IrfanView image reader. Here the big secret, invented by me 15 years ago when my consulting business was next to a print shop, is to put a matt black page behind the page you are scanning. This prevents the image on the backside of the page from bleeding through. The same Japanese engineers that take the pledge of incomplete scanning also have taken a pledge to use white platens on their scanners, when a black platen would yield better results for any page printed on both sides.

Once I get the scan into IrfanView, I use the unbelievably cool straighten tool you access by Edit>Show Paint Dialog. You don’t straighten the page by the edge, the original printer might have been on dope that day (trust me). Instead you straighten by the body of text. I do it by putting the Straighten line at the edge of a text column. Don’t be temped to align to a dingbat or a picture. The OCR needs the text straight, not the images. Once straight, I crop the image so that edges and yellowing are gone, although the auto bright setting on the Canon does do pretty good, unlike the Brother. Now you should have an image about 4600 pixels wide that you can save. That is way too big to put in a blog, so I use the IrfanView resample under the Image menu to make the page 1024px wide. Then I save that as a second file using the magnificent “save for Web” function. I go back to the high res version and reduce it to 500 pixels wide and save that as a thumbnail. I use the quality slider on the Save for Web function so that the 1024 pages are about 250kB, and the 500px wide thumbs are about 50kB. Bear in mind that scanning into a pdf does nothing but wrap your page image with a pdf file container. I am not too fond of Adobe and their bloatware, so I used gifs, jpegs, and png for web images. I just can’t understand why people want to open an image inside a pdf instead of just looking at in their browser, its all dots on a screen, after all.

You use the 600DPI for OCR (optical character recognition). I use ScanSoft I got with my Canon scanner and the PaperPort I just got with the new Bother printer. I like the PaperPort better despite its childish paradigm that you drag the images into any one of the applications it finds on your computer. When you drag the image into Word2000 it makes and rtf file and a pretty good one. Wait, I take that back. I just tried page 1 to insure that it could do 600DPI. It butchered the page lost the image and blew it out into two pages instead of one. Seamless. OK, so I used IrfanView to reduce the resolution to 300DPI and fed that into PaperPort. The Word rtf page is formatted better, but is still lost the image, perhaps because it is a bleed (fancy cool-guy printer talk for then the image goes off the page). Hang on. Yeah, OK, using the beyond fantastic IrfanView to add a 100px strip the right side of the page 1 image and now Paperport includes the image. That is not such a great thing, since Word2000 does not save images in compressed format so the file size for the single page is about 11mB, isn’t Bill Gates a visionary? I also note that PaperPort did not OCR the photo credit, that that is a horrible mistake. OK, let me add a 100px stripe to the 600DPI page 1 image. This is live blogging at its best folks. OK, adding a white stripe to the side of page 1 so there is no bleed and feeding the 600DPI image into PaperPort 12 does at least show all the text, including the photo credit. Unfortunately it blows the page out into two pages. It is also 45megB. Yeah, that’s right. So you cut the image out of the rtf file and save that and the rtf file size is 15kB. I tried for 15 minutes to get the last paragraph of the third column to jump back on the first page of this 2-page rtf, but it just wanted to hop onto a second page, no matter how I fiddled with the margins or columns. For all this seamless misery, you mind as well just send the 600DPI image to Wordpad via PaperPort and lose all the page formatting. The problem with this is that you lose all the subscripts and other UTF-8 characters.

That brings us to the images or figures on a page. Once you have the text ORC’ed, you can use IrfanView to come up with a clean small image for the figures. My big secret there is for schematics. I have found that the best results come from first reducing the size of the image with resampling. Remember that these scans are from an offset printer page, so the image is just little dots. What the reduction in size does is force a resample and the dithering algorithms inside IrfanView smooches over and smoothes out the offset printing dots. This is essential or else the figure will alias, just like an A to D converter, only in the optical realm. So then you get moiré patterns on the image, depending on what zoom you are looking at it with. So you reduce the size and the resampling takes out the offset printing dots and actually improves the image. Then you go back into IrfanView’s Save for Web. If it’s a picture you can just save the jpeg, but if it is an old EDN schematic, well this is cool, you click on a gif output, since that what you should always use for line art. Then you take the color pallet slider on the Save for Web dialog box and drag it down to 2 or 3 or maybe 5 tops, instead of 256. That crisps up the drawing and turns the mottled color op amps into a nice uniform color. It also will get the file size down to 12kB or less.

This is one nice thing about my old Scansoft OCR program. It will do stuff automatically, but it also will let you highlight parts of the page to tell the OCR algorithm whether the that section is text, an image, or a table. That tiny bit of human intervention does great, especially when you have complex formulas you want to save as a gif image. Every OCR program butchers formulas. Scansoft also lets you output your rtf as a formatted page, or as a plain page. I only wish the old Scansoft did as well with subscripts and Greek letters as does PaperPort12. I got both these for free, and the Scansoft is old, so I suspect the best thing is to spend a few hundred bucks for the newest and best OCR software. Please give me advice below if you know of OCR program that does well on technical documents like EDN.