Spark using PySpark read images

Question

Hi there I have a lot of images (lower millions) that I need to do classification on. I am using Spark and managed to read in all the images in the format of (filename1, content1), (filename2, content2) ... into a big RDD.

images = sc.wholeTextFiles("hdfs:///user/myuser/images/image/00*")  

However, I got really confused what to do with the unicode representation of the image.

Here is an example of one image/file:

(u'hdfs://NameService/user/myuser/images/image/00product.jpg', u'\ufffd\ufffd\ufffd\ufffd\x00\x10JFIF\x00\x01\x01\x01\x00`\x00`\x00\x00\ufffd\ufffd\x01\x1eExif\x00\x00II*\x00\x08\x00\x00\x00\x08\x00\x12\x01\x03\x00\x01\x00\x00\x00\x01\x00\x00\x00\x1a\x01\x05\x00\x01\x00\x00\x00n\x00\x00\x00\x1b\x01\x05\x00\x01\x00\x00\x00v\x00\x00\x00(\x01\x03\x00\x01\x00\x00\x00\x02\x00\x00\x001\x01\x02\x00\x0b\x00\x00\x00~\x00\x00\x002\x01\x02\x00\x14\x00\x00\x00\ufffd\x00\x00\x00\x13\x02\x03\x00\x01\x00\x00\x00\x01\x00\x00\x00i\ufffd\x04\x00\x01\x00\x00\x00\ufffd\x00\x00\x00\x00\x00\x00\x00`\x00\x00\x00\x01\x00\x00\x00`\x00\x00\x00\x01\x00\x00\x00GIMP 2.8.2\x00\x002013:07:29 10:41:35\x00\x07\x00\x00\ufffd\x07\x00\x04\x00\x00\x000220\ufffd\ufffd\x02\x00\x04\x00\x00\x00407\x00\x00\ufffd\x07\x00\x04\x00\x00\x000100\x01\ufffd\x03\x00\x01\x00\x00\x00\ufffd\ufffd\x00\x00\x02\ufffd\x04\x00\x01\x00\x00\x00\x04\x04\x00\x00\x03\ufffd\x04\x00\x01\x00\x00\x00X\x01\x00\x00\x05\ufffd\x04\x00\x01\x00\x00\x00\ufffd\x00\x00\x00\x00\x00\x00\x00\x02\x00\x01\x00\x02\x00\x04\x00\x00\x00R98\x00\x02\x00\x07\x00\x04\x00\x00\x000100\x00\x00\x00\x00\ufffd\ufffd\x04_http://ns.adobe.com/xap/1.0/\x00<?xpacket begin=\'\ufeff\' id=\'W5M0MpCehiHzreSzNTczkc9d\'?>\n<x:xmpmeta xmlns:x=\'adobe:ns:meta/\'>\n<rdf:RDF xmlns:rdf=\'http://www.w3.org/1999/02/22-rdf-syntax-ns#\'>\n\n <rdf:Description xmlns:exif=\'http://ns.adobe.com/exif/1.0/\'>\n  <exif:Orientation>Top-left</exif:Orientation>\n  <exif:XResolution>96</exif:XResolution>\n  <exif:YResolution>96</exif:YResolution>\n  <exif:ResolutionUnit>Inch</exif:ResolutionUnit>\n  <exif:Software>ACD Systems Digital Imaging</exif:Software>\n  <exif:DateTime>2013:07:29 10:37:00</exif:DateTime>\n  <exif:YCbCrPositioning>Centered</exif:YCbCrPositioning>\n  <exif:ExifVersion>Exif Version 2.2</exif:ExifVersion>\n  <exif:SubsecTime>407</exif:SubsecTime>\n  <exif:FlashPixVersion>FlashPix Version 1.0</exif:FlashPixVersion>\n  <exif:ColorSpace>Uncalibrated</exif:ColorSpace>\n  

Looking closer, there are actually some characters look like the metadata like

...
<x:xmpmeta xmlns:x=\'adobe:ns:meta/\'>\n<rdf:RDF xmlns:rdf=\'http://www.w3.org/1999/02/22-rdf-syntax-ns#\'>\n\n 
<rdf:Description xmlns:exif=\'http://ns.adobe.com/exif/1.0/\'>\n  
<exif:Orientation>Top-left</exif:Orientation>\n  
<exif:XResolution>96</exif:XResolution>\n  
<exif:YResolution>96</exif:YResolution>\n  
...

My previous experience was using the package scipy and related functions like 'imread' ... and the input is usually a filename. Now I really got lost what does those unicode mean and what I can do to transform it into a format that I am familiar with.

Can anyone share with me how can I read in those unicode into a scipy image (ndarray)?

Answer 1

Your data looks like the raw bytes from a real image file (JPG?). The problem with your data is that it should be bytes, not unicode. You have to figure out how to convert from unicode to bytes. There is a whole can of worms full of encoding traps you have to deal with, but you may be lucky using img.encode('iso-8859-1'). I don't know and I will not deal with that in my answer.

The raw data for a PNG image looks like this:

rawdata = '\x89PNG\r\n\x1a\n\x00\x00...\x00\x00IEND\xaeB`\x82'

Once you have it in bytes, you can create a PIL image from the raw data, and read it as a nparray:

>>> from StringIO import StringIO
>>> from PIL import Image
>>> import numpy as np
>>> np.asarray(Image.open(StringIO(rawdata)))

array([[[255, 255, 255,   0],
    [255, 255, 255,   0],
    [255, 255, 255,   0],
    ...,
    [255, 255, 255,   0],
    [255, 255, 255,   0],
    [255, 255, 255,   0]]], dtype=uint8)

All you need to make it work on Spark is SparkContext.binaryFiles:

>>> images = sc.binaryFiles("path/to/images/")
>>> image_to_array = lambda rawdata: np.asarray(Image.open(StringIO(rawdata)))
>>> images.values().map(image_to_array)