CGImageCreate Test Pattern is not working (iOS)

Question

I'm trying to create an UIImage test pattern for an iOS 5.1 device. The target UIImageView is 320x240 in size, but I was trying to create a 160x120 UIImage test pattern (future, non-test pattern images will be this size). I wanted the top half of the box to be blue and the bottom half to be red, but I get what looks like uninitialized memory corrupting the bottom of the image. The code is as follows:

int width = 160;
int height = 120;
unsigned int testData[width * height];
for(int k = 0; k < (width * height) / 2; k++)
    testData[k] = 0xFF0000FF;   // BGRA (Blue)
for(int k = (width * height) / 2; k < width * height; k++)
    testData[k] = 0x0000FFFF;   // BGRA (Red)

int bitsPerComponent = 8;
int bitsPerPixel = 32;
int bytesPerRow = 4 * width;
CGDataProviderRef provider = CGDataProviderCreateWithData(NULL, &testData, (width * height * 4), NULL);
CGColorSpaceRef colorSpaceRef = CGColorSpaceCreateDeviceRGB();      
CGBitmapInfo bitmapInfo = kCGImageAlphaNoneSkipFirst;
CGColorRenderingIntent renderingIntent = kCGRenderingIntentDefault;

CGImageRef imageRef = CGImageCreate(width, height, bitsPerComponent, bitsPerPixel, bytesPerRow, 
                                    colorSpaceRef, bitmapInfo, provider, NULL, NO,renderingIntent);

UIImage *myTestImage = [UIImage imageWithCGImage:imageRef];

This should look like another example on Stack Overflow. Anyway, I found that as I decrease the size of the test pattern the "corrupt" portion of the image increases. What is also strange is that I see lines of red in the "corrupt" portion, so it doesn't appear that I'm just messing up the sizes of components. What am I missing? It feels like something in the provider, but I don't see it.

Thanks!

Added screenshots. Here is what it looks like with kCGImageAlphaNoneSkipFirst set:

AlphaSkipFirst

And here is what it looks like with kCGImageAlphaFirst:

enter image description here

score 2 · Accepted Answer · answered Mar 26 '12 at 15:58

Your pixel data is in an automatic variable, so it's stored on the stack:

unsigned int testData[width * height];

You must be returning from the function where this data is declared. That makes the function's stack frame get popped and reused by other functions, which overwrites the data.

Your image, however, still refers to that pixel data at the same address on the stack. (CGDataProviderCreateWithData doesn't copy the data, it just refers to it.)

To fix: use malloc or CFMutableData or NSMutableData to allocate space for your pixel data on the heap.

Ah, I bet that is it. I should have realized what was happening, but I was too focused on getting the CGx functions right. I'll try to verify the answer tonight (Windows and C/C# at work, but Mac and iOS at home). Thanks! — GrandAdmiral, Mar 26 '12 at 16:49

borrrden · Answer 2 · 2012-03-26T03:15:25.873

0

Your image includes alpha which you then tell the system to ignore by skipping the most significant bits (i.e. the "B" portion of your image). Try setting it to kCGImageAlphaPremultipliedLast instead.

EDIT:

Now that I remember endianness, I realize that the program is probably reading your values in backwards, so what you might actually want is kCGImageAlphaPremultipliedFirst

edited Mar 26 '12 at 03:15

answered Mar 25 '12 at 23:54

borrrden

33,256
8
74
109

I tried your suggestion, but I still see the corruption at the bottom of the image. With kCGImageAlphaPremultipliedFirst more of the bad pixels are white in color and with kCGImageAlphaNoneSkipFirst more of the bad pixels are black in color if that helps. I also tried kCGImageAlphaFirst, but same story there. My intention is 32-bit pixels (ints) with 8 bits per color and an 8 bit alpha channel that could or could not be used. – GrandAdmiral Mar 26 '12 at 00:40
could you show a picture of the result? I am certain you want premultiplied last, so if that is not causing problems now it will later, so its best to keep it as that. Your image is BGRA, and you might need to make it as RGBA to get correct colors later but both versions have their alpha LAST not first. – borrrden Mar 26 '12 at 00:45
ps you either have alpha or you dont, you cant mix and match. – borrrden Mar 26 '12 at 00:50
I played around with "First" and "Last" and it just seemed to change the endianness of the input. I'll take alpha blending if I can get it, the real key is the ability to specify that 24-bits of color information is located in 32-bits of memory space. – GrandAdmiral Mar 26 '12 at 02:44
Command S will save a snapshot from the simulator to your desktop. You mention endian, but these constants do not have anything to do with that. You made me realize something though, since you are directly creating the data you might not be creating it with the correct endianness. Try using adding kCGBitmapByteOrder32Little or kCGBitmapByteOrder32Big to your bitmap info (bitwise OR with alpha premultiplied last). – borrrden Mar 26 '12 at 02:49
Got the screenshots now. Thanks for the tip on the Endian constants, I'll need to set that in the future when I have non-test pattern data in either format. All that said, this doesn't seem like an Endian problem. I could set the test pattern to 0x00FF0000 and 0x0000FF00 and I should just see a red/green split in some order. Unfortunately the corruption is still there. – GrandAdmiral Mar 26 '12 at 02:57
Well, I've heard rumors of this working before, try just arbitrarily increasing the size of your testData buffer to something like [width * height + 5]. – borrrden Mar 26 '12 at 03:05
Yes, I've seen the same thing in my debugging. If I make the test pattern buffer larger than it needs to be, the bitmap looks okay. Unfortunately that's probably not going to work when I have actual data. – GrandAdmiral Mar 26 '12 at 12:07

CGImageCreate Test Pattern is not working (iOS)

2 Answers2

Linked