Friday 17 July 2015

Understanding PFC3's New Autocapture Mode

The new automatic capturing mode in PFC3 may be the first that truly works as it should, but requires a little understanding to master it

Automatic scan initiation or "capturing" is an important prerequisite for a serious book scanner. With it, one has to just focus on flipping the pages and nothing else. All the machines that we know of have a button to initiate manual capturing and some of them have a timer mode whereby capture happens at fixed intervals. Timers greatly reduce the pain of having to hit a button for every single page to be scanned, especially when both hands are busy keeping the book open at the same time. However it has a drawback - you have to follow its pace. If you set the timer interval to be the shortest, your scans are done quickly but at the cost of higher stress level and you would most probably end up with a few bad scans because some pages aren't flipped in time (e.g. due to pages "sticking" together). That's when a longer interval would help but you would need to put up with its slower pace - a 2 second delay interval means you'd be spending 10 minutes longer flipping a 300-page book, and yet does not guarantee all pages can be flipped in time.

A superior alternative to the timer mode is the automatic capturing mode (ACM). This is how it's supposed to work - it scans instantly by itself when a page is flipped and ready, but it waits when it's not. Sounds wonderful doesn't it? But wait - actual implementation and realization of this wonder is a tough job, even in this day and age of advanced computing. The machine needs to know when a page is "flipped and ready". The "ready" part is relatively straightforward, but the problem is the "flipped" part - it doesn't take much for a human to tell that a page is flipped but for a machine? Not easy. The computer algorithm either needs to somehow "identify" the page flipping process and/or check that the page content has changed. Checking page content is definitely the easier way of the two, but it still requires a significant amount of processing time - that means the ACM would no longer react instantly, and worse still, it may not even react at all if it makes a mistake in judgment (i.e. when current page content is similar to the last). That's frustrating, and we have evidenced this behavior in at least one scanner costing twice the xcanex. Of course, the algorithm can be made so advanced that it never makes a mistake but in doing so, it would be too slow that one is better off using the timer mode.

That explains why we did what we did for the ACM algorithm in PFC3 - it no longer waits for a page flip and will snap instantly when there's an open book in the scan area (either single or dual page). What it does instead is detect for motion - once there's no motion for a split second, it'll capture. The advantage of this over the timer mode is that you can take your time to flip a page in case pages stick and it'll wait for you so long as there's motion (i.e. keep your fingers moving). If you want it to pause, simply shift the book slightly out of the preview window. So do not leave an open book fully inside the scan area without any motion - it'll continue to snap away. Alternatively, you can do it the conventional way - just hit the "A" key on your keyboard to toggle in and out of auto mode.

We understand the risk of designing the ACM to behave this way. Users may expect it to wait for a page flip after every capture before snapping the next, and gets startled when it doesn't. But we feel that users would prefer the advantage of having instant capture to minimize scanning time, and forgo the need for it to check for a page flip which incurs a delay during every capture. However to use it effectively, users need to understand its behavior - the new ACM will capture instantly when an open book is within the scan area plus no motion is detected, regardless of whether there has been a page flip. We think the lack of page-flip checking is a non-issue (once understood, that is) and users would rather have a much faster and smoother ACM. But do let us know if we're wrong and it turns out to be a problem for you.

Another improvement in the new ACM is its Autofocus (AF). There are now 2 types of AF - the normal and fast AF. The normal AF takes a few seconds to sweep the whole focusing range to cater for a possible step change in book thickness (e.g. focal point changes substantially when a 200-page book is swapped with a 1000-page book). On the other hand, the fast AF assumes that contiguous pages are being scanned from the same book, hence it just takes a split second to make minor adjustments to the focal point. This is how the new ACM uses these 2 types of AF: if a page is flipped swiftly right after a capture and the subsequent page is set and ready within a few seconds, fast AF shall be used and the capture is almost instant. However if a longer period of motion is detected (more than 3~5 seconds) before a page is ready, the normal long-range AF is used instead in anticipation of a larger adjustment in focal point due a user switching the book or skipping a large chunk of pages. So when you are scanning contiguous pages, immediately flip to the subsequent page after a capture and set it down swiftly - this will activate the fast AF and saves considerable amount of time when scanning larger volumes.

The video below shows a total of 6 captures made using the new ACM - in the first 3 captures, extended motion was intentionally applied to show how ACM would wait for the page to be set and ready before it automatically initiates the scans. The last 3 captures demonstrates how ACM reacts instantly when pages are flipped immediately after each capture and set swiftly for the subsequent scan.




It is my hope that the new ACM will improve your work efficiency, and I'm always happy to hear your feedback!

Tuesday 14 July 2015

perfecapture3: Five Years of Incremental Innovation

Release of PFC3 for the xcanex bookscanner

Faster, better scan quality, intelligent new features, smoother operation, increased efficiency and more robust than ever


 The new perfecapture3 
     
Building a good book scanner is so much tougher than we thought – and maybe that’s why book scanners today are either too expensive or not very useful. Some problems we face seem impossible to have answers but we know there is simply no shortcut – again and again we must study, think, create and experiment. Supposedly hours of work stretch into days, days into weeks, and weeks into months just because a desired result stubbornly avoids realization. We carry on because our mission is to create the world’s best book scanner for the masses and not anything less. To compromise is to defeat the purpose. So is the goal reached? We await your answer.

Cut the crap and give it to me


Ad-hoc footage of PFC3 undergoing final tests

Summary of changes (from v2.114)

Twice the capture speed: Added intelligence and faster autofocus reduced capture time from 6 to 3 seconds (We define ‘capture time’ as the time a page is set for capture until it’s done and can be flipped). Structural optimizations at the core of the PFC3 created the smoothest operation we’ve ever seen when scanning in large volumes.

Ultraflat text: Many page flattening algorithms produce slight waviness in text sentences after processing but we cannot settle for that. We demand scans that appear indistinguishable from flatbed scans. The new Flatpage v3.0 algorithm now analyzes page content, measures remaining waviness and makes critical adjustments to produce the finest looking paragraphs ever from a platen-less book scanning system.

Hardcover book mode: Automatic cropping and digital page flattening of hardcovers (or hardbacks) are tough jobs for software algorithms. Most scanning apps require either manual adjustments or book inserts to produce good scans. A breakthrough in our Enhanced Edge Detection algorithm changed that – any hardcover can now be scanned as easy as a paperback.

xcanex with buddies: The mountain of books used to evaluate the PFC3

PerfectPage Technology: Even after typical digital corrections to book curvature, scans made without using cumbersome book clamps/platens and without any manual post-scan corrections still suffer inherent inferiority due to distortions caused by content shifts, page dents and twists. That’s why all those manual work, clamps and platens still prevail despite the suffering they bring to people. So we cracked our brains and created PerfectPage - it works in conjunction with Flatpage and intelligently performs content centering (i.e. intelligent means it shouldn’t center what’s not supposed to be at the center) and uncompromisingly eliminates all those stubborn distortions. PerfectPage + FlatPage technologies = no more platens and manual work.


Inferior no more

10x reduction in hard disk space requirement: Thanks to advanced data compression methods, it now requires less than 5GB of hard disk space to scan 500 pages at one go. This is a big relief for smaller solid state drives.

A better automatic capture mode: Ever felt that the Timer capture mode (TC) is too demanding and the Auto capture (AC) is too sluggish? The problem with TC is that it doesn’t wait for you and the drawback of AC is that you have to wait for it. So we decided to improve the AC mode to capture instantly when you’re ready but pauses when a page is being flipped or when the book is slightly shifted out of the scan area. Learn more about it here.

Improved finger image removal: We realized that quite a lot of finger images were not automatically removed on scans of certain books. It took us a while to figure out that the culprits were page dents (due to finger pressure on softer paper) and extreme coloration (caused by paper aging). Adding intelligence to tackle them wasn’t easy, but we succeeded and the occurrence of unremoved finger image is now less than 1% (typical) on these books.

Tweaked user interface: Visual aids were added to tell users exactly when capture happens so that the page can be flipped in preparation for the next capture. The confusing capture mode icon (A/M and T/M) is now more intuitive and we got rid of that irritating translucent band jitter in the preview. The PFC3 also allows you to scroll through your scans by just clicking on any thumbnail and then tapping the up or down arrow keys.


Well that’s about it. If you have pre-installed perfecapture on your PC, just download and run the patch and it will be upgraded to PFC3 within a minute. Download the patch or full installer here.

We want to hear from you. Let us know your thoughts about PFC3 by writing to us at yourfriend@piqximaging.com or via our contact form.




Thanks.
piQx Team