Google’s new PhotoScan app and the trade-off of quality vs time

Yesterday Google launched PhotoScan, a new app to help people bring their old printed photo collections into the digital world faster and more efficiently.

PhotoScan is an app that was made for people like me: proud archivists of their family’s old photos. We who spend hours, days, weeks trying to find, scan, categorize and share these photos. Google said PhotoScan “gets you great looking digital copies in seconds – it detects edges, straightens the image, rotates it to the correct orientation, and removes glare.” Google adds that it also saves the scanned photos to Google Photos “to be organized, searchable, shared, and safely backed up at high quality.” So, scan and organize are the goals. Great.

A few years ago I used an old flatbed scanner to scan all the photos of my grandparents and their families to ensure that they were preserved forever and that all cousins had digital copies. I went through about 200 photos and for each I did a pre-scan, adjusted the scanning parameters to expand the contrast range and a pre-crop (i.e. scan only the photo, not the entire page) and did a final scan. Each photo took me about 5-10 minutes to get right, because most were small. In my collection, around 70% were 2” by 3”, 25% were 3” by 5”, and 5%, mostly group photos, were 5” by 7”. I’ll get to the significance of these stats later.

Today I downloaded the PhotoScan app and played around with it. It is extremely easy to use and, relative to a flatbed scanner, really, really fast! Here’s how it works:

  1. Position the photo in the frame. This works even if it’s placed carelessly, crooked, with other photos on the page, or behind a glass or plastic film.
  2. Align the phone with the four corner dots in the app. PhotoScan then processes the photo for a few seconds.
  3. If necessary, adjust corners by pulling the four corners of the 4-sided outline to match the photo.

That’s it, the photo is saved directly to Google Photos.

Position, click at four corners, and adjust corners if necessary. Note the tiny zoom + cross-hair aid for adjustment.

Position, click at four corners, and adjust corners if necessary.
Note the tiny zoom + crosshairs aid for adjustment.

If PhotoScan manages to grab the photo correctly, the overall time to “scan” a photo is around 30 seconds. If the corners need to be adjusted then it takes closer to a minute. This is because the adjustment is frustrating to do with touch control. Even though the app provides a zoom in and crosshairs for each corner as it’s grabbed, this is a task best done with taps as opposed to drag. It’s just too sensitive and each movement too large. This makes the entire corner-adjusting process too slow and frustrating, especially when scanning multiple photos. That said, the image processing to detect, straighten, unskew, and “rectangularize” a photo is amazingly good. Also, if the photo is alone and better situated (i.e. fully frontal camera view) in the initial frame then PhotoScan detects the photo correctly more often.

My test photo: PhotoScan's version.

My test photo/

This brings us to the big trade-off: time vs quality. The photo I used to test PhotoScan is small, 3.5” by 5. When I scanned it on my flatbed scanner, it came out as 15.1MB bitmap. When I converted it to high-quality JPG with Photoshop, it was 3.58MB. The advantage of this size is that every fraction of the photo was represented, down to the texture of the paper it was printed on. Since I scanned at such a high resolution, I could then enlarge the photo and show it on a big screen and print it out at double or even quadruple its original size and it still looked good. For group photos, the higher resolution allowed me to zoom in on the faces, which was an advantage. PhotoScan saved this image as a 293KB JPG.

Now, you could argue that the time saved is worth the reduction in quality. Optimally the scan time is 30 seconds on PhotoScan versus 5 minutes on a flatbed, even though this was a few years ago and I assume today the process would be faster than that, but I digress. You would be right if you had thousands of photos to scan. The use case, however, tells a different story.

Photos from the early half of the last century were rare and infrequent. My grandparents and their ancestors didn’t own a camera, they waited for a traveling photographer to reach their town and take a handful of photos of them, their family, and their friends. In the second half of the century, when personal cameras were more prevalent, the limitations of a roll of film, with either 24 or 36 photos, and the hassle and cost involved in developing and printing a roll limited the number of photos. I’d be curious to find a stat but even in the 1980s taking more than, say, 10 photos in a single event was extravagant and rare. For my project, 200 photos spanned 6 decades. So, while it can be daunting to look at entire albums and shoeboxes of photos, there just weren’t that many, especially in the first part of the century. Even if using PhotoScan is faster, with the truly important photos the lower quality is a deterrent. After all, the goal is to do it once for posterity, right? Quality is important.

My second gripe is organization. We rely a lot on Google’s amazing photo recognition skills, and they are usually mind-blowing. Every time Google is able to recognize faces and differentiate between siblings across time, I am amazed anew. Yet right now, PhotoScan saves a photo without any info. For my test photo, it named it with a meaningless sequence of 45 letters and numbers whereas I had named it with the name of people in it and the year in which it was taken. It would be great if PhotoScan popped up a quick dialog to enter date and the names of people in the photo at least. That way it would also have more information to detect the faces other photos.

Front and back of a photo of my great-aunt and her husband. On the back a dedication to her aunt, their names, the date and the occasion, all in Polish.

Front and back of a photo of my great-aunt and her husband.
On the back a dedication to her aunt, their names, the date and the occasion, all in Polish.

My third gripe, though of lesser importance than the first two, is only for the crazy archivists that like to scan the back of every historical photo as well as the front. This is because there is usually a lot of important information on it such as the subjects of the photo, the occasion and date it was taken, and often reason for sending it. Some of my favorite discoveries have been love poems behind a few innocent-looking photos. There needs to be way to tie the two scans of the front and back as one photo.

Note: on the photo on the right, the text on the back was written in Polish, which I do not understand. It took a few weeks for me to find a speaker who had the patience to untangle the handwriting and translate for me. Sadly she couldn’t decipher all the words. This would be something Google AI would be great at!

Finally, would I use PhotoScan and not my old flatbed scanner if I was doing the project I described above today? Probably not, just because of the quality issue. Sure, I’d save time, but we’re talking about saving these relatively few photos for future generations. Were tagging and organizing given more attention, I’d reconsider, but as of now, quality wins.



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s