[whatwg] Image resize API proposal

This really seems like the wrong solution. Taken to an extreme next you'll
need to add VideoRisizer, AudioRecompresser, and any thing else JavaScript
can't do without freezing the browser.

It seems like it would be better to figure out a way to get Web Workers to
be able to do this. Even if they have to XHR the binary down, decompress
into a TypeArray (see WebGL) and read the data themsevles so they can keep
the EXIF stuff, bloating the browser for one small use doesn't seem like the
right solution.

You can post an open source library and the usage can be as simple as this
proposal.


On Tue, May 11, 2010 at 11:58 AM, Sterling Swigart <sswigart at google.com>wrote:

> I'm working with David Levin, and based on the feedback received regarding
> offscreen canvas, the proposal has been changed to address more specific
> scenarios. The main use case was resizing images, so we are proposing an
> asynchronous image resizing API. If you are curious about how we arrived at
> our API below, take a look at the "appendix" to view the alternatives we
> considered.
>
> Let us know what you think. Thanks!
> Sterling
>
> Use Cases:
>
> Begin with a user giving a local image file to a webpage. Then:
>
> 1. In real-time chat, quickly give other users a thumbnail view of the
> image file.
>
> 2. Or, limit the size of an image file before uploading it to a web server.
>

This use case is already handled (minus the EXIF).

> Proposed Solution:
>
> We propose adding image.getBlob. getBlob will be an instance function of
> the javascript Image object which asynchronously gets a blob of the image,
> resized to the given width and height, encoded into jpeg or png. The
> function declaration will be:
>
> getBlob(mimeType /* req */, width /* req */, height /* req */, successEvent
> /* req */, errorEvent /* op */, qualityLevel /* op */, preserveAspectRatio
> /* op */, rotateExif /* op */);
>
> The blob will be passed as an argument to the success callback function, or
> upon error, error data will be passed into the error callback function as an
> argument. Quality level should be between 0.0 and 1.0, and any value outside
> of that range will be reverted to the default, 0.85. If MIME type does not
> equal "image/jpeg", then quality level is ignored. If null (or a negative
> value) is passed in for the width or height, then the function will use the
> source's measurement for that dimension. Default values for
> preserveAspectRatio and rotateExif are true.
>
> All EXIF metadata will be retained except for any saved thumbnails, and the
> EXIF rotation property will be appropriately modified.
>
> Security:
>
> If the image source is of a different origin than the script context, then
> getBlob raises a SECURITY_ERR exception.
>
> Sample Code:
>
> // url contains location of an image file
>
> Image i = new Image();
>
> i.src = url;
>
> var successEvt = function (newBlob) { myDiv.innerHTML += "<img src='" +
> newBlob.url + "' />"; };
>
> var errEvt = function (err) { alert(err); };
>
> i.getBlob("image/jpeg", 300, 350, successEvt, errEvt, .55);
>
> // Image will retain aspect ratio and correct for EXIF rotation. If the
> source image was 700x700,
>
> // the blob will represent a new image that is 300x300.
>
> That's all!Appendix: Alternatives considered
>
> For reference, we've also included a list of other designs that we thought
> of along with the reasons why they were dropped
>
> Creating a new object for resizing
>
> Summary of approach:
>
> [NamedConstructor=ImageResizer(),
>
> NamedConstructor=ImageResizer(blob, onsuccess),
>
> NamedConstructor=ImageResizer(blob, onsuccess, onerror),
>
> NamedConstructor=ImageResizer(blob, onsuccess, onerror, type),
>
> NamedConstructor=ImageResizer(blob, onsuccess, onerror, type, width,
> height)]
>
> interface ImageResizer {
>
> void start(); // starts resize operation
>
> void abort(); // aborts operation
>
>  attribute Blob blob;
>
> attribute DOMString type; // default "image/png"
>
> attribute unsigned long width;
>
> attribute unsigned long height;
>
> attribute float qualityLevel; // default 1.0, must be 0.0 to 1.0, else
> reverts to default
>
>  readonly attribute unsigned short started; // default 0
>
>  attribute Function onsuccess;
>
> attribute Function onerror;
>
> };
>
> Why it wasn't chosen:
>
> Creating an entirely new object for this task made the task seem more
> complicated and involved than necessary, and this problem could be solved
> via modifications to the Image object.
>
> Returning a SizelessBlob immediately from a method on image
>
> Summary of approach:
>
> var streamingBlob = image.toStreamingBlob(mimeType /* req */, width /* req
> */, height /* req */, qualityLevel /* op */, preserveAspectRatio /* op */,
> rotateExif /* op */);
>
> New Blob Interfaces:
>
> interface SizelessBlob {
>
> // moved from Blob
>
> readonly attribute DOMString type;
>
> readonly attribute DOMString url; // whatever name -- URL, urn, URN, etc.
>
> }
>
> interface StreamingBlob : SizelessBlob {
>
> // at most one of the following functions will be called for a single
> FutureBlob
>
> attribute Function onblobready;
>
> attribute Function onerror;
>
> readyonly attribute blob; // throws an exception if accessed before
> onblobready is called.
>
> }
>
> interface Blob : SizelessBlob {
>
> readonly attribute unsigned long long size;
>
> Blob slice(in long long start,
>
> in long long length); // raises DOMException
>
> };
>
> Why it wasn't chosen:
>
>    - the disconnect of the error from the thing that caused it making
>    failures hard to understand (e.g. An image load may fail but that may not be
>    detected until the xhr is done using the resized image.)
>    - the issues that result from passing a SizelessBlob, which has a
>    reference to a loading image, to another document and closing the original
>    document (thus killing the image loader)
>    - introduction of multiple blobs which may be confusing to developers
>    - the need to change all existing specs to use SizelessBlob instead of
>    Blob
>
> Returning a Blob immediately from a method on image
>
> Summary of approach:
>
> var blob = image.toBlob(mimeType /* req */, width /* req */, height /* req
> */, qualityLevel /* op */, preserveAspectRatio /* op */, rotateExif /* op
> */);
>
> blob.size and blob.slice throw until the blob is complete. There would need
> to be a new event to say when the blob's size is ready.
>
> Why it wasn't chosen:
>
> It felt like it was changing how blobs worked to make the size throw.
> Also, this has some of the same disadvantages of as the SizelessBlobapproach.
>
> Using CSS of Image to designate dimensions instead of putting it in the API
>
> Summary of approach:
>
> img.getBlob(mimeType /* req */, successEvent /* req */, errorEvent /* op
> */, qualityLevel /* op */);
>
> Why it wasn't chosen:
>
> It would have been confusing to the user if getBlob took into account some
> CSS attributes but not others, and using all CSS tags posed a lot of
> unnecessary implementation complexity without any use cases.
>
> Using the File object
>
> Summary of approach:
>
> FileReader readResized(). The result would be a data url with the resized
> image.
>
> Why it wasn't chosen:
>
> We also examined making this function an instance method of the FileReader
> object, but this function did not fit well in the context of the File
> object. FileReader's result is a string, posing a problem for large results
> as well as an unnecessary conversion. One way another this problem is to use
> a URN (like File.urn), it would have prevented us from posting the result to
> a server as well as having the lifetime issues that File.urn does.
>
> Having a very specialized method on FormData to do the resize and allow
> upload
>
> Summary of approach:
>
> FormData.appendResized(DOMString name, Blob value, mimeType /* req */,
> with /* req */, height /* req */, qualityLevel /* op */,
> preserveAspectRatio /* op */, rotateExif /* op */);
>
> Why it wasn't chosen:
>
> This approach was over specialized and didn't allow for using the result on
> the web page at all which would be useful as a preview of exactly what is
> being uploaded.
>
> Using canvas in a worker
>
> Summary of approach:
>
> This is the OffscreenCanvas proposal which can be seen here: h<http://www.mail-archive.com/whatwg at lists.whatwg.org/msg20297.html>
> ttp <http://www.mail-archive.com/whatwg at lists.whatwg.org/msg20297.html>:<http://www.mail-archive.com/whatwg at lists.whatwg.org/msg20297.html>
> //www.mail-archive.com/whatwg at lists.whatwg.org/msg20297.html<http://www.mail-archive.com/whatwg at lists.whatwg.org/msg20297.html>
>
> Why it wasn't chosen:
>
> Due to making it rather complicated for the javascript user to accomplish
> their goal and the code complexity and the speed trade-offs that may change
> due to gpu acceleration, this idea didn't get traction and many folks
> suggested something simpler for the intended use cases. A few more drawbacks
> are that this doesn't obey exif automatically and it doesn't preserve the
> exif metadata.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.whatwg.org/pipermail/whatwg-whatwg.org/attachments/20100513/e8215ffd/attachment-0001.htm>

Received on Thursday, 13 May 2010 15:32:27 UTC