Re: Any idea to make zxing work with blurry image?

[Sean Owen]

Yes, here are the things to try, in order of increasing difficulty and probable benefit:

1. Smarter edge detector, like a modified Canny detector. This gives moderate but noticeable improvement

2. If you know pretty well the point-spread function (like, you know that the image will be at the same distance and so will blur in the same way) you can make a deconvolution kernel to recover something more like the original first. This doesn't help if you don't know the optics, but probably helps a lot if you have a fixed device and fixed frame to scan

3. If you can put the barcode reliably in the same region of the image, for 1D barcodes, then you can learn to read the blurry image by learning something like a neural network over training examples of digits. 

The problem is that 2 doesn't work for the general case of scanning on many different phones at many distances. 3 doesn't work for 2D barcodes and doesn't work if you can't force the barcode to be about the same size each time in the image. (But of course for the common case of 1D product codes, where you force the image to be in a fixed part of the view finder, it probably works well.)

I haven't bothered with 3 because most devices have auto-focus, it doesn't help all formats, and well it's hard enough work. But that is definitely what I'd try if it were important.

[CGR]

Do we have any solution from ZXing library to decode the barcodes from the blurry image (taken from iPad2) ?

Thanks