do-it yourself repro v-cradle for paper books

[ereszet]

Originally Posted by user

mm, RemoteCapture would be so important reason to buy a camera? ...I recognize that all these would be handy ofcourse, but would it justify a compromise in image quality?

Sorry for the poor quality photo of the laptop screen in my previous post. I forgot about PrintScreen. So here you are (pictures attached).

Previewing a photo on the laptop screen and changing zoom and other settings "on the fly" to get the photo right is crucial for me. Resolution is important but lighting and focus "make or break" the photo. You cannot see the quality of the picture in the camera viewer.

If you are cost conscious you have to suffer compromises.

[ereszet]

It is time for examples of quality photo scans. For that, I took the photo of a page from an Ukrainian album on Kiev (Kyiv). Please note, that this is only to illustrate some procedures and therefore for copyright reasons I spoiled the photo with an overlayed text. The album is larger in size than a home scanner can take, so the photo scanning is the only way to copy the page. The photo scanning setup was not ideal due to a temporary arrangement that I use now. In particular, the lighting was a 50 watt overhead halogen bulb, ceiling lamps switched on, and a lighting diffusor. An ideal setup would require a bright (possibly special photo reprographic) overhead lamp.

In the first attached picture (image processing.jpg) you can see the original photo, the same photo binarized to black and white, and then the same binarized photo with the color picture inserted. To achieve that result you just copy the picture from the color original and paste it into the binarized one. You can do it with the free Irfanview program.

Why should one use color pictures in black and white pages instead of dealing with original color pages? For two reasons. First of all, a book of several hundred pages scanned in color will produce a set of images/pdf file of enormous size. Secondly, the black print in color pages is not quite black and the white background is not quite white. For comparison see the attached pictures (aleksander ocr pdf color.jpg and aleksander ocr pdf bw and color.jpg). The three images of OCRed photos in both pictures are sequentially the following pdf formats: text under picture, text over picture, and text and picture only.

Finally, I attach printscreens (lrf color vs black and white with color picture.jpg) from Sony Reader's Connect software that show lrf pages respectively for color and binarized (with color pictures inserted) pdf pages (text under image) refined with pdflrf program. Although the large size album with small print is not something you would carry with you in the Sony Reader, this example shows vividly the diiference between rendering color and mono pages in the Reader's lrf pages.

[ereszet]

Long overdue Finereader 9 has been released. I will test it and report the results in this thread as soon asr I buy it. In the meantime I quote below some hints regarding photo scanning from:
http://abbyy.com/DLCenter/downloadce...de_English.pdf


1. Make sure that the page fits entirely within the frame.
2. Make sure that lighting is evenly distributed across the page and that there are no dark areas or shadows.
3. Straighten out the page if required and position the camera parallel to the plane of the document so that the lens looks to the center of the text being photographed.

Minimum Requirements
. 2–megapixel sensor
. Variable focus lens (fixed–focus cameras, common in cell phones and hand–held devices, will usually produce images unsuitable for OCR)
Recommended Requirements
. 5–megapixel sensor
. Flash disable feature
. Manual aperture control or aperture priority mode
. Manual focusing
. An anti–shake system, otherwise the use of a tripod is recommended
. Optical zoom

Shooting Modes

Lighting
Make sure there is enough light (preferably daylight). In artificial lighting, use two light sources positioned to avoid shadows. (ereszet remark: this is not necessary with the v-cradle design; one bright overhead light source and a light diffusor is recommended).

Positioning the Camera
If possible, use a tripod. Position the lens parallel to the plane of the document and point it toward the center of the text.
At full optical zoom, the distance between the camera and the document must be sufficient to fit the entire document into the frame. Usually this distance will be 50–60 cm. (ereszet remark: tripod with a ball head is not preccise enough to set the camera angle and distance to match different size paper documents every time they change; repro copy stand with micrometer plates will save you a lot of positioning effort; with a tripod, once you position the camera, do not touch it anymore, do all the distance and angle positioning with the v-cradle instead).

Flash
Whenever possible, turn off the flash to avoid glare and sharp shadows on the page. In poor lighting conditions, try using the flash from a distance of about 50 cm, or, preferably, use additional lighting.
Important! Using the flash when photographing documents printed on glossy paper causes the worst glare. (ereszet remark: see my post on redirecting the flash to the ceiling; glare does not interfere with the shots taken with the v-cradle setup).

White Balance
If your camera allows, use a white sheet of paper to set white balance. Otherwise, select the white balance mode which best suits the current lighting conditions. (ereszet remark: white sheet of paper may not work in bright sun; see an example in one of my previous posts; try Kodak grey card instead).

Auto focus may not work properly in poor lighting or when photographing at a close distance. In poor lighting conditions, try using an additional light source. When photographing a document up close, try using the Macro (or Close–Up) mode. Otherwise, if possible, focus the camera manually (ereszet remark: set the zoom and preview the focus with PSRemote remote capture program; if a page is almost blank with a few lines of text use another page of text to set the focus and fix the focus.)
If only a part of the picture is blurred, try reducing the aperture value. Increase the distance between the document and the camera and use maximum zoom. Focus on a point anywhere in between the center and a border of the image (ereszet remark: as long as that point is not a blank part of the page).

[user]

Question:
I wonder now if we really need so many megapixels or its useless or overkill. And if not, what is the most importan factor to achieve highest OCR rates.

Scanner:
The above is a scan. I dont know how this scan was produced, what scanner was used, what scanner settings were used, I only know its dimensions (A4), its resolution (850x1103, 100dpi) and that its black-and-white.



Here is the scan that was OCRed:

Here is the OCRed text:

Finereader results:
Uncertain characters: 18
Total characters: 2087
OCR success rate: 99.14%

The Finereader results were superb in my opinion, considering that, as you can see, most of the uncertain characters were some music characters.

This results make me really wonder. In order to produce a 100dpi image of A4 size you only need 0.9MP. So there must be something more than just resolution. Abby cannot say "you just need a 300dpi image" in order to OCR successfuly.

Camera solution:
The Fujifilm F31fd has excellent sharpness and image quality and 6.3MP resolution. It will produce very sharp images of 3024x2016 resolution and as the article from this respected website says, "Fuji has managed, with (F31fd's) sensor and processor combination, closer than ever before to 'SLR-like' output from a compact camera, when its compared with the SLR Nikon D50 (6.24MP, 23.7 x 15.6 mm sensor, +1k USD)

Images of 6.3MP resolution in the A4 size produce 260dpi, which is below Finereader's minimum requirements, but as I said, I wonder if this dpi is really not enough.

Finereader has 300dpi as the only (afaik) requirement that the image needs to have in order to perform accurate OCR.
Obviously, Finereader can't "see"/OCR too big fonts or too small fonts in the image and there must be an optimum font size or, better, range that OCR success rate is highest. But why it doesnt have specific pixel size, brightness, contrast, etc in the minimum requirements for accurate OCR?

In my opinion, Finereader not only needs the font in the captured image to be of specific size, but of specific quality as well (sharp image with good contrast, smooth boundaries of the lines, etc). We need to know more parameters in order to achieve 100% OCR rate (99.2% OCR rate is not enough, because it means 600 mistakes that need to be manually corrected in a 300 pages book, thats looks like a lot of job)

[user]

Font size or Font quality or both?

Note:
document = the page of the book with the printed text
image = the capture of the document, either with a scanner (scan), either with a camera (photo)

Font size on the image:
The font size in the image depends on the font size in the document and the resolution that we use to capture the image.

Font size on the document:
It can be measured in points
1 point = 0.3527 mm
Usualy fonts are 5-22p in printed text pages

As bigger the font is, the lower resolution is needed to produce the same font size in the captured image (I suppose we can make a graph or the ratio for "image resolution with which the page is captured" (x) and "font size in the printed page" (y)).
For example (I dont know the exact amounts), if a 20p font needs to be captured in 5MP resolution in order to have the optimum for OCR size in the image, then a 10p font needs to be captured in 10MP resolution in order to have the optimum for OCR size in the image.

So the optimum resolution varies and depends on the font size in the document. But why should we adjust resolution? Because we need the minimum resolution that gives the best OCR rate, because cameras with many megapixels have alot of noise and this is affecting the quality of the image (shooting in bigger resolution than needed, not only is overkill, not only is a waste of money for expensive camera, but it's also known that when MP increase, IQ decrease... etc+).

So we should shoot or scan to specific resolution according to the font size in order to achieve
however maybe there are not much differences between different shooting/scanning resolutions for normal and usual (10-20p) fonts. Maybe there is no advantage. Maybe the different resolutions havent so big difference (eg 0.0x megapixels). Also, maybe Finereader "normalizes" resolutions according to font size. These are just some thoughts to increase OCR accuracy and to find the ultimate camera for the job.

I wonder if Finereader needs more megapixels because it only needs big fonts, or it needs more pixels and better pixel quality as well, or all of these.

If Finereader needs more megapixels just for making fonts bigger, we should examine magnification.
One method is shooting at high resolution, so if we increase the resolution the image is magnified (shooting in 10MP and then in 12MP, the image size is increased by 10%).

Another solution is that if we have a low resolution captured image, we can magnify it with an image editing program in order to make the fonts in the image bigger.
So would only magnification be sufficient for achieving optimum font size in the image?
No, because image editing software magnification results in distortion of the image (and thus fonts), because it doesnt add pixels, it just magnify the current pixels:

So we need software that will trim megapixels and produce accurate lines and circles, and the trimming will be in proportion with the given character line of the specific font type and size.
One method to achive this is interpolation. Any photoshop guru may help us for magnify-without-distorting methods, and any other software or interpolation methods to do this.

Another thing to consider is that fonts are usualy of specific type, eg arial, verdana, etc.
I dont know if modern books mention somewhere which font are typed in.
I also dont know if there is a way (a program, or some rules) to identify by ourselfs the font type that a book is typed in.

But if we know the font type and the font size, I suppose it would be easier to tell the OCR program what to compare with (since it should have the whole character set of the specigic font type in the specific size).

I suppose and hope Finereader already does automatically all of the above (identifies font type, font size, etc and compares to the given font type all the characters line, interpolates if necessary, adjusts contrast, brightness, etc)
But, Abby's support is fast, though abysmal and not technical at all, afaik, and Finereader doesn't come with more technical requirements in order to achieve fine images and hit high OCR rates. So I would really like to know if Finereader doesnt do any of the above mentioned solutions so that we can help it more.

[user]

I just shoot a book page with my 2MP camera of my cellphone

the only problem I encountered is that corners are blurry and dark and there is a distortion as the book page doesnt lay flat

I am quite satisfied with it, I am almost sure that we don't need too many megapixels, but sharp lens (no blurry corners, barrel distortion and dark corners) and as less as possible page curve

the flat page problem:
a glass over a bookpage doesn't seem to flaten it enough
I wonder how the kirtas machine flatens the book pages (or how it corrects the distortion with a postprocess operation)
what I think it does is that it shoots the photo when the bookpage is vertical:

this is the best position that the page is completely flat and there is no distortion at all

if we manage to reproduce this, to make some kind of photocopy stand that will be able to stabilize each bookpage in that position, the main (since all others seem easily correctable) problem of book scans will be solved

EDIT: some programs are supposed to fix the curve distortion of book pages, but I havent tried any yet (if needed I 'll post some links)

EDIT2: if we put a single book page between two glasses we will do the job nicely, however it will be very time consuming to do this for every page

but only books with soft cover have problem to be flaten, the hard cover books are easier to flat

[ereszet]

OCR success rate of 99% is acceptable in a number of applications (like book scanning) and unacceptable in other applications (like bank documents).
For a "paperless office", which today in most cases is based on picture true copies and some keywords describing the documents, an index of 99% properly recognized words is more than one could wish for (and some of the remaining 1% words are just uncertain - not necessarily errors). For special applications (like musical notes) there are special OCR programs.

OCRing books is important for two different reasons:
either to save the result in a variety of text formats (i.e. Word or html) in order to work on the text later (e.g. to convert to lrf text format)
or to save the text under the page image in order to have a perfect copy of the original and to index the words.

In the first case 1% error rate does not prevent you from reading the book. As a matter of fact, the Discovery TV channel proves in their spots that you need only the first and last letters of the word in context to make the right associations in your brain. Gutenberg project books (see Websters encyclopedia of 1911) contain a lot misrecognized letters but are usable anyway.

In the second case, you are able to search thousands of books for some combination of words, even using wildcards, e.g. you can search for all the pages with words "photo scanning" or "photo* scan*") and the displayed pages will be exactly like the original. If you miss a page or two because of ocr errors, you can live with it. I use it for genealogical reseach even with 17th century books (where e.g. "s" is printed as "f"), and I can still find a lot of references to what I search for.

As for the ideal parameters for scanning before OCR, there is simply no such requirement or need. Twenty years ago, when OCR was based on simple pattern matching, there were very strict rules for the resolution and recognizable fonts. Today, OCR algorithms are much more sophisticated and can cope with a plethora of languages and font shapes. However, they cannot make an informed decision, whether an object (like a big title or a small icon or a musical note) is a picture or text. Therefore, warnings and errors are unavoidable. I stress it again that even the perfect 300 dpi picture of the text you convert from Word directly (without printing to paper and scanning) will have OCR errors and warnings. The only way to improve on that is to teach (train) the OCR program with the first few pages of the book, so that it can easily tell the difference between e.g. "n" and "h" in the fonts used to print the book.

Quality of scans (focus, uniform lighting, proper positioning) is more important than the resolution, whether you plan to OCR the scan or not. Finereader can cope with resolutions from 96 dpi up. The new Finereader 9 can automatically set the proper brightness of greyscale images for scanning but for the camera you need to set the parameters yourself. Because every book is different you should always take some trial photos of a few pages at different camera settings and preview the results (you may scale the picture up to see the details). If your eye is happy with the results, your OCR program should be happy as well.

Finally, as books are concerned, scanners are no match to photocopying, because of limited size, slow operation, deformation of scanned book pages, damage to books, etc.

BTW. While Finereader 9 allows for 2 Mpixels camera shots, it specifically says in its guide that cellphone cameras are not fit for the purpose. I wonder why you try all kind of approaches instead of using a solution with which I started this thread, and which I tested on tens of thousands of book and document pages. Just try it, even with the "quick and dirty" cardboard v-cradle.

[user]

I am just trying to find the optimum camera for the job

firstly, I was about to buy an expensive 12MP camera, but now I am quite sure its overkill, MP is not an important factor (since even my 2MP cellphone camera does the job)

megapixels myth seems to have effect in OCR as well

I wonder if a good image quality 12MP camera will do it better or worse than a 6MP best best image quality. Also I wonder which specific criteria for image quality are most important for OCR (noise, artifacts, bluriness, dark corners):

http://bpribadi.multiply.com/photos/photo/13/1
http://bpribadi.multiply.com/photos/photo/13/2

also, I am trying to make a set of rules to increase OCR rate. For example, if Finereader usually mistakes n with h in a specific document, it should ask us to train it.

[ereszet]

Originally Posted by user

I am just trying to find the optimum camera for the job

You get what you pay for (camera) but the best things in life are free (DIY v-cradle). When you buy a camera after a long research, the next day you learn about a better one.

Some cameras come with a text/document mode. Do not fall for it but just note what are the settings they provide for that mode. E.g. my Casio EX-Z850 uses the following settings: EV Shift +1, white balance Auto, ISO Auto, Sharpness +1, Saturation 0, Contrast +1. Of all these parameters the only one worth considering when shooting a document in automatic mode is EV Shift (exposure compensation), which provides for brighter pictures. The other parameters can be adjusted by software in processing stage (modifying sharpness before OCR is not recommended). With my rather dim lighting setup I usually use EV between +1 and +2. Of course, when I set the parameters manually (e.g. ISO 50, Tv 2 seconds, Av 3.5) there is no need for exposure compensation.

As for the camera resolution, there is little room for myths. The higher is the resolution the camera sensor supports (without overcrowding the pixels) the better are the details of the picture. You need the details when shooting a document with very small print. Although small print comes usually in small size printed objects (like business cards) and the camera can be zoomed to fill the frame with the picture of the object (which means dpi equivalent of 600 or even 1200 dpi), there are some publications of bigger sizes with very small print.

An example of such a small print are instructions for the use of medicines provided by pharmaceutical companies which are difficult to read even with the strong glasses. I took pictures of such a leaflet at two different zoom levels to see what difference the magnification makes to the OCR. The higher is the zoom the higher is the dpi equivalent. The OCR error/warning rate changed from 2% to 1% with the two times higher zoom. So the resolution matters.

[ereszet]

Originally Posted by user

what I think it does is that it shoots the photo when the bookpage is vertical...this is the best position that the page is completely flat and there is no distortion at all...if we manage to reproduce this, to make some kind of photocopy stand that will be able to stabilize each bookpage in that position, the main (since all others seem easily correctable) problem of book scans will be solved

You are trying to reinvent and complicate the wheel. To shoot a photo of a book page in vertical position you would have to place the camera parallel to the page while the other pages obstruct the view (as your picture of Kirtas repro stand shows). And no, page curvature is not a problem at all in a v-cradle with adjustable angle wings. You can set one of the wings almost vertically with the other one almost flat. The most difficult thing is to ensure uniform lighting, and that is what v-cradle helps to achieve. In most cases, the best angle between the v-cradle panel wings is 120 degrees.

[user]

can you tell me please if there are other cameras apart from canon that can be remotely controled from the pc?

thanks

PS: unfortunately I got an email from Sony that says cybershots dont have such feature

[ereszet]

Originally Posted by user

can you tell me please if there are other cameras apart from canon that can be remotely controled from the pc?

Visit this page:
http://www.steves-digicams.com/digso...ameractrl.html

Also, search that site for "Camera Control Software" to see discussions at some forums.

For Canon models see:
www.breezesys.com

I am aware of some Olympus cameras and some high end Nikon cameras, but I am quite happy with my Canon and PSRemote by Breezesys (original Remote Capture by Canon is no match for PSRemote), so I have not followed the subject for some time.

[Studio717]

My Olympus C-7070 WZ works with the software Cam2Com, though it's PC only.

My particular camera model has been discontinued, but it's still listed as working with the software. I'd recommend finding the software you want to use, and then seeing what cameras are able to use it, if you're going to be buying a new camera anyway.

[user]

thanks for your replies

I don't know which software I need, I suppose any software that can control the camera from the pc

however, I think camera needs to support such function as well

I am about to buy a Casio, Fuji, Pentax or Sony, not interested for others

[ereszet]

I am afraid, you are out of luck, and none of those cameras can be operated remotely from the computer. If you want to stay with that choice, you should look for a camera with at least an infrared remote.