Use member-uploaded covers before Amazon when adding from library sources

I understand that adding books from amazon will include an amazon cover, but why are amazon covers automatically used when adding via OverCat? The best-guess algorithm seems to manage very well if there is no amazon cover, so why not give it priority over amazon when using library sources?

I've been re-entering a lot of my amazon records using OverCat, and it's really annoying when there is a high-quality member-uploaded cover (which I uploaded myself!) but the system still defaults to amazon.

So I recommend that the choice of covers for library sources be changed to prioritise high quality member-uploaded covers over amazon covers.

Taking this from the other topic:

For any ISBN, LibraryThing goes through a cascade of "is it there?" failure:

1. A cover you selected or uploaded for that book
2. Amazon
3. Member covers from someone else--most popular one for that ISBN.

I don't think swapping 3 and 2 would actually be a good idea. But I could be persuaded out of it.

What would be the negatives to swapping 3 and 2?

> 3

For one, member covers from someone else are more likely to be bad quality. I'd rather have an Amazon cover than a blurry, unreadable one, and I've seen plenty of member-uploaded covers like that.

Perhaps, but I'd rather have a bad quality member-uploaded cover that is an easy reminder to change it than one from Amazon (which would get switched eventually anyway, and is just as likely to be the wrong cover altogether).

#4
I hate blurry covers too, lilithcat - that's why I said 'high-quality' covers in my suggestion. That would appear to need an extra step if Tim's 2 and 3 were swapped: some kind of filter that would ignore covers without the 'high-quality' label.

(moved from the other topic)

2> Correct me if I'm wrong, but wouldn't 3 only be hit a tiny tiny fraction of the time? Just curious, but it would seem that way to me, due to Amazon's wide coverage of ISBNs.

On the other thread, lorax replied to #7 with:

In your library, the "member uploaded, best guess by ISBN" option accounts for 124 covers.

( http://www.librarything.com/profile/brightcopy/stats/covers )

I think somethings a bit screwy there. I clicked on a few of those, and the ISBNs were definitely found at amazon.

8>

Yes, quite weird. Take this for example. If you enter 0312852770 directly into amazon.com, it finds the book and shows a cover.

If you add 0312852770 on LT using amazon as a data source, it gets a result but does not show an amazon cover. Instead, it shows the third member-uploaded cover (not even the high res one).

Very peculiar.

>9 brightcopy:

Sometimes Amazon.com has a cover, but they don't show it outside of Amazon. For example, that Harvest of Stars book has one on Amazon, but they don't share it via their cover service.

Here's the URL to their covers:
http://images.amazon.com/images/P/0195153855.01._SX140_SY225_SCLZZZZZZZ_.jpg

Change the 0195153855 to the ISBN10 of any item. In this case, the URL goes to a 1x1 transparent pixel—Amazon's sign that they don't, in fact, "have" the image. Or they have it but won't share. Either way, no cover.

I think this usually means the image is a "customer image." But not always, I think. Maybe they have some covers from somewhere with no permission to show them elsewhere. I dunno.

Occasionally LT doesn't know Amazon has the cover. We don't check every time. We check once in a while. If that returns no cover or a blank cover, we mark it in the database as a no. Sometimes that "no" stays too long. Sometimes Amazon has it, but one call out of a 1,000 goes afoul and we mark it "no" wrongly.

'll confess to being really clueless as to what it takes to make these things happen, but here goes anyway.

Would it be possible that when a book is added a window, box, whatever pops up and asks if we want to change the cover?

Or better still, could change cover be added to the quick edit thats part of recently added?

Yes, I've thought about that, especially when:

1. You enter a book without an ISBN, and
2. There are other non-ISBN books in that work to which a cover has been added

Even adding it to the main 'edit' page would be a huge help.

#2
I don't think swapping 3 and 2 would actually be a good idea. But I could be persuaded out of it.

1. I thought you were intending to reduce LibraryThing's reliance on Amazon - this seems helpful towards that goal.

2. Using Amazon covers leaves that annoying pitfall where the Amazon link may change the ISBN without the user realising. Most people who choose library sources are likely to care more than average about the accuracy of their data, and finding out that an undetermined number of their records may have been changed seems to make them rather cross.

3. While I go to the trouble of scanning covers mostly for my own benefit, I like to think that the high-quality scans I've done are contributing to the general quality of LibraryThing. Knowing that the system will only use them as a last resort is a bit of a let-down. (I know you don't keep records of who uploaded which cover, but a 'Your scan is now the primary image for ISBN X' running count (and badge!) would be cute, and probably encourage scanners.)

4. Using the Amazon cover leaves a permanent Amazon link on that record, even when it's been replaced by a member-uploaded cover. I gather that for people who have completely switched their records from amazon, this can be a nails-on-the-blackboard kind of irritation.

>14 EveleenM:

1. Yes, it would be helpful, especially if it encouraged more uploading of covers.

2. Fair enough, but having LT show the user cover would be FAR more likely to produce change. Part of the point in the LT cover is that we use the majrity one. (I think we should also take into account image quality.) But anyway, that will change over time--probably more than the Amazon one. Freezing it in place doesn't attract me very much.

Also, even if we did freeze it, doing it now would turn people's covers upside down. We'd have death threats. In many cases we'd take a perfectly good Amazon cover and replace it with something blurry, or a scan of a cover with a ding in it, or whatever.

3. I think MOST cover uploading is for books that have no covers. Doing that is obviously a big help.

Note that when we go mobile, we'll show only user covers.

4. That won't be changing anytime soon. There's a lot of Amazon data here and there. We'd need to rewrite all sorts of data structures so that they noted what was and what wasn't, and knew the bottom part of the page when it made the top, etc.

#15
2. Fair enough, but having LT show the user cover would be FAR more likely to produce change.

I don't quite understand you here: I was referring to changes made to people's data without them realising it; I think that's pretty different from people choosing to change stuff themselves.

Also, even if we did freeze it, doing it now would turn people's covers upside down. We'd have death threats. In many cases we'd take a perfectly good Amazon cover and replace it with something blurry, or a scan of a cover with a ding in it, or whatever.

I'm not sure where freezing it comes in; it seems reasonable enough that the most commonly picked high-quality image may change over time. And with reference to the blurry part, I did specify that this would need some kind of filter to make sure that the quality of the image on offer was up to scratch. With the size of the files you now accept for upload, I think the most recent member-uploaded covers are often better quality than the Amazon versions.

I think MOST cover uploading is for books that have no covers. Doing that is obviously a big help

I think it varies by genre; most of my SF and fantasy works had member-uploaded versions to match all the Amazon covers, while more of the mysteries had no member-uploaded versions.

>16 EveleenM:

I think we're both missing each other. You are arguing that Amazon imges are bad because Amazon may change the image that corresponds to an ISBN. I'm pointing out that LT does that too. If you upload your own image, fine. But if you use an ISBN for which LT provides the cover, that cover can change as we get better covers and as people use them.

#17
What I was arguing is that Amazon can change the ISBN, which I think is more of a problem for people who are picky about their data. (At any rate, that's what I've been told - is this another case where the common wisdom is wrong?)

I think it only changes the ISBN if you pick an Amazon cover that isn't linked to the ISBN in your catalogue.

So if your entered ISBN is 0123, and the cover you've chosen uses ISBN 2468, then the ISBN in your catalogue changes to 2468. But if you don't change anything and it goes with best guess, then it doesn't change your ISBN, but you're also more likely to see the cover change as the best-fit ISBN's cover changes.

I could be wrong, and it does change your ISBN even without you choosing an Amazon cover, but I don't think that's the case.

I've changed the "confirm" popup. It now says

Choose this cover? This will change your book's ISBN.

If you choose an Amazon ISBN. Otherwise it says

Choose this cover?

I think it's better now, but it makes perfect sense for the cover and the ISBN to stay in synch. There are some exceptions, but mostly not.

Thanks, Tim, I think that change will help clarify things.

it makes perfect sense for the cover and the ISBN to stay in synch

Except that there may be multiple editions with different ISBNs but the same cover (or at least similar -- for example differing only in the unreadable blurb at the bottom or top). Just because I've chosen the cover that is the closest match to mine doesn't mean that the ISBN will match mine.

Could it be a bit more explicit and say "Choose this cover? Choosing an Amazon cover will change your book's ISBN; choosing a user-provided cover will not."

Also, if you're avoiding automatically using user-provided covers, you could keep a count of how many times each user-provided cover is being used (which you do, if you delete it when the use count gets to zero, right?). When someone adds a book, choose the most-used user cover for other books that match their ISBN (if there is one).

It would be a big help if we could tell which cover we actually have for that book. (Comparing the pictures often isn't enough.)

Since people have very different feelings about using Amazon covers, and even changing Amazon covers to default covers leaves them labeled 'Amazon covers', I think it would be good if there were a choice in the account settings that let us decide whether or not we want Amazon covers.

15> 2. Fair enough, but having LT show the user cover would be FAR more likely to produce change. Part of the point in the LT cover is that we use the majrity one. (I think we should also take into account image quality.) But anyway, that will change over time--probably more than the Amazon one. Freezing it in place doesn't attract me very much.

What?! That's completely counter-intuitive. None of the other data in your catalog record can change without your knowing about it (well, except for ISBN which has been discussed above). If one adds a book, glancing at the cover and seeing that it matches your edition, I think very few users will think that it might change at a later time.

#24
I agree, most people check the data when they enter it, and don't expect it to change again without their own intervention. So I certainly think that best-guess covers should be kept to the exact image originally entered.

When you add a book, you shouldn't get the "best guess cover for the work, which can change over time" or even the "best guess cover for the ISBN, which can change over time" -- you should get a particular cover, which is at that moment the most popular high-quality user provided cover for the ISBN, and it should not change after that unless explicitly changed by the user, regardless of whether the relative popularity or any other factor changes.

26> Yes, very much so. This is a concept which should really be seriously considered. If I don't pick a cover and leave the one that was auto-picked, this should be treated like I picked the one that was auto-picked. I can see possible downsides to that, but I think the upsides trump it.

24/25/26> me too :)

Hear, hear.

Yes yes yes!

I'm in support of *anything* (well, almost) that helps get more member-covers, and helps lessen reliance on Amazon covers.

(oh, and just fyi, the "high quality" marker on member-uploaded covers doesn't always appear, sometimes doesn't appear for a LONG time after uploading the cover, so I'm not sure we can rely just on that as an indicator of quality. Also, I *have* seen some fairly blurry "high quality" ones.)

If you *really* want to increase the number of member covers (which would be a significant asset that you could sell to LTFL libraries), then I suggest the following:

(1) Suggest images based on publisher/ISBN data. (What I mean is that if you know the publisher and the ISBN, you can make a pretty good guess as to the image URL on the publishers website. For example, Random House is http://www.randomhouse.com/catalog/covers_450/9780307266101.jpg, OUP is http://ukcatalogue.oup.com/images/en_US/covers/large/9780192729927_450.jpg) The images could be suggested on the 'change cover' page, or on one of the pages mentioned below.

(2) Allow users to mark duplicate images. If users marked a user-cover as a duplicate of an Amazon cover, you could effectively retire that particular Amazon cover. You might want a quality threshold (at least xx pixels, and possibly a certain number of votes). Ideally we ought also to be able to merge user-uploaded covers, so newer high-quality images can be chosen in place of poorer quality ones (though obviously the user who uploaded the image would still get his own, regardless of merging).

(3) Award badges, so users know you want the images.

(4) Create a page which shows all your books that have no images, but where images (Amazon or user) exist. Then allow users to select their preferred cover in one click from this page. It would mean we could add covers to hundreds of books in a few minutes.

(5) Similarly, create a page which shows all our 'guessed' covers, where we can confirm them with one click (although I'm sure some of my 'guessed' covers are ones I've actually selected).

The advantage to LT of providing tools that help us select the best image (even if we don't actually upload something new) is two fold:

(a) The more covers are chosen by users, the better data you'll have about which covers match which books.

(b) By allowing members to quickly choose correct covers, we'll find it much easier to see which books would require us to actually scan and upload covers. If I knew (for example) that I only needed to scan 100 covers to de-Amazon my library, I'd be much more motivated to do that

31.2>

I really don't like this idea. Different people's ideas of "identical" and "duplicate" vary widely. I want the exact cover that my edition has -- title in the same color, no extra "badges" describing award-winning status or "special introductory price" -- in good condition -- minimal scratches and creases, and no price tags or other damage. Other people will consider anything that has approximately the same cover art to be "duplicate", and vote accordingly, while still others will be happy only with a scan of their own cover, with their own scratches, dings, and price tags. No voting system on "duplicates" can possibly address this.

31.5>

This exists. Look at your Statistics/Book Covers page, and select "user-uploaded, best guess by ISBN." This provides you with a view in your library of all those "guess" covers.

I'm probably imagining this, as I'm prone to wishful thinking about LT features I desire, but over the past several days the books I've added to LT have been automatically assigned to member-uploaded covers -- despite matching the existing amazon cover for the same ISBN.

Has anybody else noticed this happening?

I think allowing us to "label" member uploaded images with an ISBN would be helpful. Then LT could have member uploaded cover suggests that aren't dependent on solely on popularity.

LibraryThing already knows the ISBN you've entered for your book, and the cover you've chosen, so it can make that connection. It shouldn't be too difficult for the system to use data it already has to choose the most popular user-uploaded cover for a particular ISBN.

@lorax (32)

Regarding duplicate covers. I'm not suggesting that you can't upload your own unique cover if that's what you want. But I'm suggesting that by default duplicate covers are hidden from users.

Regarding the page with our guessed covers. I'm aware of the page you refer to, but as I said, I specifically wanted to be able to confirm all these covers with one click. I want to move covers from 'guessed' to 'selected' in one click.

36>

You are, however, suggesting that instead of being able to piggyback on other people having uploaded the exact cover I want to use, I need to upload my own or risk having it overridden by people who decide that something that looks vaguely the same if you squint at it without your glasses on is a "duplicate".

#36
I'm not suggesting that you can't upload your own unique cover if that's what you want. But I'm suggesting that by default duplicate covers are hidden from users.

I see two problems with this:
Covers which look the same at first glance, or in catalogue view, but are actually different on closer inspection. The same basic picture may have a different placing of text, an added blurb, or an overprinted prize or series label: a lot of people would consider them the same cover; others (like me) would only be happy with a cover which is absolutely identical.

Secondly, a very high-detail scan may have a brightness, contrast or colour balance that other users don't like (maybe due to varying monitor settings). And the many covers which have gold or silver metallic inks can produce huge variations in brightness depending on the scanner - which version a user prefers is very much up to individual taste.

For example, have a look at http://www.librarything.com/work/7313/covers/61020026. It takes a bit of thought to work out which of those covers are really identical; then I think which of them looks closest to the actual cover will depend very much on the settings of your monitor.

So while I still feel, as in the original post, that high-quality member-uploaded covers should be used, I don't think the choice of uploaded covers should be restricted.

On the flip side, look at http://www.librarything.com/work/2773690/covers ; wouldn't it be nice to have some way of digging through that? By default duplicate covers are hidden from users doesn't restrict the choice of anyone. In the case you're looking at, even if you do want to be picky about your cover, would it be a big deal to pick which of the seven covers is approximately yours and then sort through the specific covers that are similar, instead of digging through all 15 covers?

It may be semantics, but to my mind "group similar covers together, and collapse the groups by default" is very different from "merge duplicate covers". And, yes, "merge", not "hide", is the word used in the post I was originally responding to; the poster may have changed his mind, which is good (and may now attempt to reflect that by changing history, as so often happens in the name of "clarification", so I'll quote it here:)

Ideally we ought also to be able to merge user-uploaded covers, so newer high-quality images can be chosen in place of poorer quality ones (though obviously the user who uploaded the image would still get his own, regardless of merging).

That's what I am opposed to, and it's a far cry from keeping all covers and just grouping them by similarity.

Let me clarify my thinking on 'merge' vs. 'group'. (I happily confess my thinking is developing here. That's the point of a forum, isn't it? I have no urge to edit an earlier post to make myself look 'right' for posterity.)

I still think that Amazon images should be merged (i.e. permanently hidden) when an identical user-uploaded cover is added, assuming the quality is adequate. Anyone who chooses an Amazon cover does so with the risk that the image will be replaced at some point. In that situation, I really can't see how anyone can complain if it's replaced by an almost identical non-Amazon cover.

In some situations it would be better for non-Amazon covers to be also merged (I'm trying to avoid situations like this: http://www.librarything.com/work/256698/covers/61618170), though I accept Lorax's arguments that it wouldn't always be best. An automatic grouping, with the option to show all similar results would perhaps be the best compromise. (Doing for covers what Overcat does with editions.)

The grouping with the option to show all does sound promising. I understand not wanting to see, say, 25 member-uploaded covers when 15 are almost exactly the same as one another.

However, as an avid cover-uploader, I will upload my own cover even if there is an exact member-uploaded match, because my specific cover may have a tiny tear on the bottom or a sticker on the top, and I want *that* cover representing my book in my catalogue. Not everyone is as finicky as that, though.

42> So in that case, you don't need to look at the covers at all, since you want _your_ cover. Sounds good to me!

I was very ambivalent on merging covers, for many of the reasons elucidated above (especially what EveleenM said), but grouping by similar covers would be wonderful. I frequently find a long list of high-quality covers (especially on 'classic' works) and have to scroll and scroll to find a cover that a) matches mine and b) is high-quality to MY mind. If I could look at a) first and separately, then search only among those for b), I would be delighted!

Oh, Tiiim! Someone came up with a leap for your new image method to expand to...

I would definitely like a way to group similar covers. It could use the split author interface?

Since many cover differences are related to edition, one of the best ways to differentiate covers will probably be the publication date. Of course then, this information will need to be tied to each cover image (daunting!). Once this data is available, when you go to choose a cover, those with the date matching your book info can be shown first with the option to view all.

This or some kind of textual information needs to be added (rather like Cover Guess is doing really) as computers/scripts really cannot visually recognize things that are obvious to our human eyes in order to organize or group them.

What we need is a heraldic language for book covers: A maid with argent mane, rampant, upon a azure field.

ETA: Replaced or with argent because people would think or was a typo.

I wonder if the algorithm that the reverse image search (http://www.tineye.com) uses would be useful here, or if because the images often come from different scans rather than being manipulated versions of the same original image, it wouldn't recognize the elements as being the same. (I confess that I'm not entirely clear on how tineye's search process works.)

46> Is that example from an sf novel, then?

Yes, the grouping would have to be automated for it to be successful. There's freely available software which should be relatively easy to integrate into LT if Tim wanted to. http://libpuzzle.pureftpd.org/project/libpuzzle

#47: Tineye is pretty good about recognizing different scans; I've seen it match a crop of a black and white painting to a book cover on Amazon that had a different crop of the same painting in color. It's overkill for LT's purposes, I think; something like scaling to 50 pixel height and only matching if the least squares of the differences are below a certain level should work.

findimagedupes does (from the manual)

To calculate an image fingerprint:
1) Read image.
2) Resample to 160x160 to standardize size.
3) Grayscale by reducing saturation.
4) Blur a lot to get rid of noise.
5) Normalize to spread out intensity as much as possible.
6) Equalize to make image as contrasty as possible.
7) Resample again down to 16x16.
8) Reduce to 1bpp.
9) The fingerprint is this raw image data.

To compare two images for similarity:
1) Take fingerprint pairs and xor them.
2) Compute the percentage of 1 bits in the result.
3) If percentage exceeds threshold, declare files to be similar.

It works passably.

49> That's cool! I did some exploration of Tineye a while back, but it looks like it's more powerful than it was then. Dunno how long it's been since I looked at it, but the website looks a lot different now.

Mostly, though, I was thinking that Tineye already has an algorithm that could be used to group cover images as requested, and if it works - and according to the findimagedupes that you've copied it would - then there's little to stop from implementing something similar for works with >x number of covers (well, processing power and servers and whatnot would stop it, and the need to recalculate however often, but that's beside the point).

With books like Pride & Prejudice or the Wizard of Oz with over 400 covers, it'd be awfully useful.

Can I throw in a request for Check #4 up there in Message #2 with a "Grab most used cover"? I know with my old RPG material, Amazon doesn;t have a cover for the work, an ISBN's never been entered but yet 2 or 3 if not more covers have already been uploaded. And yet the blank cover is what gets assigned.

This happened a lot on the Dragon Magazine stuff. (Which was strange as ISBNs are in there for those.)

How does the option "Automatically show member-contributed book covers" on the Edit profile and settings/Miscellaneous effect the covers shown?

As I'm entering my Tolkiens atm, my hands are itching to do something with the 400+ images on each book. Group them by language, then edition, would be so nice. And then combine some which are so obviously the same, and you can just flat out remove a couple of really bad blurry ones which have perfectly good version, exactly the same.

Bump.
Just look at this one: http://www.librarything.com/work/578489/covers
There's a beautiful high-resolution member-uploaded cover, used four times, but the lone Amazon cover attached to an incorrect ISBN gets chosen, and can't be flagged.

Yes, please. LT is good in crowd sourcing, so why not use it for this kind of thing.