Question

I’m trying to do a mock-up in Matlab on some image data in a database. The image data is from java, output as a base64 encoded byte array. I’m not familiar with the [java image] format. However, I wrote the following Matlab code based on some java written by someone working with me. They follow the same basic outline, and the Java code is able to read the image just fine. The Matlab code looks like this:

function [ result ] = queryDb( theQuery )
  conn   = database( ... ); % connect to the database
  result = fetch( exec( conn, theQuery ) );
  result = result.Data;

  close( conn );
end

data = queryDb( 'sql query to get the data' );
data = uint8( data{1,1} );
data = org.apache.commons.codec.binary.Base64.decodeBase64( data );
data = uint8( 127 + data ); % the base64 decoder returns signed int8

import javax.imageio.ImageIO;
import java.io.ByteArrayInputStream;

dataStream    = ByteArrayInputStream( data );
bufferedimage = ImageIO.read( dataStream );

Upon inspection of bufferedimage, it is an empty array of double as opposed to a java BufferedImage instance.

I ran a few tests on dataStream to see if it behaved as expected; sort of a mini unit test along the lines of:

for jj = 1:10
  for kk 1:10
    assert( dataStream.read() == data(kk) );
  end;

  dataStream.reset();
end;

It checked out, so this leads me to believe the problem is with ImageIO or my use of it.

Unfortunately, none of the examples I’ve found for using ImageIO (and some of these other APIs) are used in quite the manner I outline here (in Matlab that is).

This code uses java.io.ByteArrayInputStream in the same manner — in the sense that the provided data is an array of bytes.

This code is essentially what I’m looking to do — convert a java image to a Matlab array. Unfortunately, they cheat by taking a Matlab image, turning it into a java image, then turn it back.

This code uses ImageIO, but it does so by reading from a file stream. I tried writing the data out to a file then reading it in using java.io.File, but I get the same result either way.

So, I’m at a loss.

Answer 1

The approach outlined in the question works … the only thing I have that differs in any way from the code above is I’m not adding 127 to the uint8 array:

% Do all the imports
% sql query to get the data
dataStream    = ByteArrayInputStream( decodeBase64( rows.data{1} ) );
bufferedImage = ImageIO.read( dataStream );
[w h] = deal( theWidth, theHeight );
imageData     = uint8( bufferedImage.getData.getPixels( 0,0,w,h,[] ) );
imageData     = permute( reshape( imageData, [3 w h] ), [2 3 1] );

imshow( imageData );
% voila

---

edit Incorporating the text @kchja linked to, in case that link becomes invalid:

Subject: How can I convert a “Java Image” object into a MATLAB image
matrix?

Problem Description: The IM2JAVA function is available for converting
a MATLAB image into a Java image. I would like to convert a Java image
to a MATLAB image.

Solution: There is no built-in function to convert Java images into
MATLAB images. However using the Java API you can extract the data
from a Java image and store it as a matrix, which is MATLAB’s image
representation.

Below is an example of converting a MATLAB image to Java and back to a
MATLAB image. The image used in this example is part of the MATLAB
demos and is on the MATLAB path. In this example, the “getPixels”
function of the Java class “Raster” returns the RGB values for the
image.

Depending upon the format of your Java image, additional work may be
required. Java images can be stored in one of many formats; the
“getPixels” function returns the pixel data in that format. For more
information, see the javadoc page for the java.awt.image.Raster class
at http://java.sun.com/j2se/1.5.0/docs/api/java/awt/image/Raster.html

% create java image
I = imread('office_3.jpg');
javaImage = im2java(I);

% get image properties
H=javaImage.getHeight;
W=javaImage.getWidth;

% repackage as a 3D array (MATLAB image format)
B = uint8(zeros([H,W,3]));
pixelsData = uint8(javaImage.getBufferedImage.getData.getPixels(0,0,W,H,[]));
for i = 1 : H
base = (i-1)*W*3+1;
B(i,1:W,:) = deal(reshape(pixelsData(base:(base+3*W-1)),3,W)');
end

% display image
imshow(B);

The following are two other ways to implement this (in a more
optimized manner):

Example 1: (Faster execution)

pixelsData = reshape(typecast(jImage.getData.getDataStorage, 'uint8'), 4, w, h);
imgData = cat(3, ...
        transpose(reshape(pixelsData(3, :, :), w, h)), ...
        transpose(reshape(pixelsData(2, :, :), w, h)), ...
        transpose(reshape(pixelsData(1, :, :), w, h)));

Example 2:

imgData = zeros([H,W,3],'uint8');
pixelsData = reshape(typecast(javaImage.getBufferedImage.getData.getDataStorage,'uint32'),W,H).';
imgData(:,:,3) = bitshift(bitand(pixelsData,256^1-1),-8*0);
imgData(:,:,2) = bitshift(bitand(pixelsData,256^2-1),-8*1);
imgData(:,:,1) = bitshift(bitand(pixelsData,256^3-1),-8*2);