I'm trying to create image filter that will shift color of the image. In order to do this I need to convert rgb color to hsl and after shift, convert hsl back to rgb. I make some researches and found formulas that can help me with this task.
I implement them in my playground using Swift just to test if they are reliable and they are. I won't post Swift code here just to keep things clean, but I'll show my test results:
input: rgb (61, 117,237) or (0.24,0.46,0.93)
result:
rgb2hsl [0.613527 0.831325 0.585] or (221, 83, 58.5) //hsl
hsl2rgb [0.24 0.46 0.93] //back to rgb
Great! So far so good.
Now we need to convert our Swift code to Core Image Kernel Language (CIKL).
And here it is:
float hue2rgb(float f1, float f2, float hue) {
if (hue < 0) {
hue += 1.0;
}
else if (hue > 1) {
hue -= 1.0;
}
float res;
if (6*hue<1) {
res = f1 + (f2 - f1) * 6 * hue;
}
else if (2*hue<1) {
res = f2;
}
else if (3*hue<2) {
res = f1 + (f2 - f1) * (2.0/3.0 - hue) * 6;
}
else {
res = f1;
}
return res;
}
vec3 hsl2rgb(vec3 hsl) {
vec3 rgb;
if (hsl.y == 0) {
rgb = vec3(hsl.z,hsl.z,hsl.z);
}
else {
float f2;
if (hsl.z < 0.5) {
f2 = hsl.z * (1.0 + hsl.y);
}
else {
f2 = hsl.z + hsl.y - hsl.y * hsl.z;
}
float f1 = 2 * hsl.z - f2;
float r = hue2rgb(f1, f2, hsl.x + 1.0/3.0);
float g = hue2rgb(f1, f2, hsl.x);
float b = hue2rgb(f1, f2, hsl.x - 1.0/3.0);
rgb = vec3(r,g,b);
}
return rgb;
}
vec3 rgb2hsl(vec3 rgb) {
float maxC = max(rgb.x, max(rgb.y,rgb.z));
float minC = min(rgb.x, min(rgb.y,rgb.z));
float l = (maxC + maxC)/2.0;
float h = 0;
float s = 0;
if (maxC != minC) {
float d = maxC - minC;
s = l > 0.5 ? d / (2.0 - maxC - minC) : d / (maxC + minC);
if (maxC == rgb.x) {
h = (rgb.y - rgb.z) / d + (rgb.y < rgb.z ? 6.0 : 0);
} else if (maxC == rgb.y) {
h = (rgb.z - rgb.x) / d + 2.0;
}
else {
h = (rgb.x - rgb.y) / d + 4.0;
}
h /= 6.0;
}
return vec3(h,s,l);
}
And here comes the problem. I'm not able to get right values using this functions in my filter. To check everything I made a Quartz Composer Patch.
Since I didn't find any print/log option in CIKL, I made this to check if my conversions work right:
The logic of this patch: my filter takes color as an input, convert it to hsl and back to rgb and returns it; image input ignored for now.
Kernel func of my filter:
kernel vec4 kernelFunc(__sample pixel, __color color) {
vec3 vec = color.rgb;
vec3 hsl = rgb2hsl(vec);
return vec4(hsl2rgb(hsl), 1);
}
Filter includes functions listed above.
The result I see in the viewer is:
Image on the right is cropped constant color image from the input color.
The left image is the output from our filter.
Digital color picker returns rgb (237, 239.7, 252) for left image.
I have no more ideas how to debug this thing and find a problem. Any help will be highly appreciated. Thanks.
I found the problem. It was me, converting code from Swift to CIKL I made a stupid mistake that was very hard to find, because you have no print / log tools in CIKL or I don't know about it.
Anyway, the problem was in the rgb2hsl function:
float l = (maxC + maxC)/2.0; // WRONG
it should be:
float l = (maxC + minC)/2.0;
Hope it'll help someone in the future.
I want to convert from cube map [figure1] into an equirectangular panorama [figure2].
Figure1
Figure2
It is possible to go from Spherical to Cubic (by following: Convert 2:1 equirectangular panorama to cube map ), but lost on how to reverse it.
Figure2 is to be rendered into a sphere using Unity.
Assuming the input image is in the following cubemap format:
The goal is to project the image to the equirectangular format like so:
The conversion algorithm is rather straightforward.
In order to calculate the best estimate of the color at each pixel in the equirectangular image given a cubemap with 6 faces:
Firstly, calculate polar coordinates that correspond to each pixel in
the spherical image.
Secondly, using the polar coordinates form a vector and determine on
which face of the cubemap and which pixel of that face the vector
lies; just like a raycast from the center of a cube would hit one of
its sides and a specific point on that side.
Keep in mind that there are multiple methods to estimate the color of a pixel in the equirectangular image given a normalized coordinate (u,v) on a specific face of a cubemap. The most basic method, which is a very raw approximation and will be used in this answer for simplicity's sake, is to round the coordinates to a specific pixel and use that pixel. Other more advanced methods could calculate an average of a few neighbouring pixels.
The implementation of the algorithm will vary depending on the context. I did a quick implementation in Unity3D C# that shows how to implement the algorithm in a real world scenario. It runs on the CPU, there is a lot room for improvement but it is easy to understand.
using UnityEngine;
public static class CubemapConverter
{
public static byte[] ConvertToEquirectangular(Texture2D sourceTexture, int outputWidth, int outputHeight)
{
Texture2D equiTexture = new Texture2D(outputWidth, outputHeight, TextureFormat.ARGB32, false);
float u, v; //Normalised texture coordinates, from 0 to 1, starting at lower left corner
float phi, theta; //Polar coordinates
int cubeFaceWidth, cubeFaceHeight;
cubeFaceWidth = sourceTexture.width / 4; //4 horizontal faces
cubeFaceHeight = sourceTexture.height / 3; //3 vertical faces
for (int j = 0; j < equiTexture.height; j++)
{
//Rows start from the bottom
v = 1 - ((float)j / equiTexture.height);
theta = v * Mathf.PI;
for (int i = 0; i < equiTexture.width; i++)
{
//Columns start from the left
u = ((float)i / equiTexture.width);
phi = u * 2 * Mathf.PI;
float x, y, z; //Unit vector
x = Mathf.Sin(phi) * Mathf.Sin(theta) * -1;
y = Mathf.Cos(theta);
z = Mathf.Cos(phi) * Mathf.Sin(theta) * -1;
float xa, ya, za;
float a;
a = Mathf.Max(new float[3] { Mathf.Abs(x), Mathf.Abs(y), Mathf.Abs(z) });
//Vector Parallel to the unit vector that lies on one of the cube faces
xa = x / a;
ya = y / a;
za = z / a;
Color color;
int xPixel, yPixel;
int xOffset, yOffset;
if (xa == 1)
{
//Right
xPixel = (int)((((za + 1f) / 2f) - 1f) * cubeFaceWidth);
xOffset = 2 * cubeFaceWidth; //Offset
yPixel = (int)((((ya + 1f) / 2f)) * cubeFaceHeight);
yOffset = cubeFaceHeight; //Offset
}
else if (xa == -1)
{
//Left
xPixel = (int)((((za + 1f) / 2f)) * cubeFaceWidth);
xOffset = 0;
yPixel = (int)((((ya + 1f) / 2f)) * cubeFaceHeight);
yOffset = cubeFaceHeight;
}
else if (ya == 1)
{
//Up
xPixel = (int)((((xa + 1f) / 2f)) * cubeFaceWidth);
xOffset = cubeFaceWidth;
yPixel = (int)((((za + 1f) / 2f) - 1f) * cubeFaceHeight);
yOffset = 2 * cubeFaceHeight;
}
else if (ya == -1)
{
//Down
xPixel = (int)((((xa + 1f) / 2f)) * cubeFaceWidth);
xOffset = cubeFaceWidth;
yPixel = (int)((((za + 1f) / 2f)) * cubeFaceHeight);
yOffset = 0;
}
else if (za == 1)
{
//Front
xPixel = (int)((((xa + 1f) / 2f)) * cubeFaceWidth);
xOffset = cubeFaceWidth;
yPixel = (int)((((ya + 1f) / 2f)) * cubeFaceHeight);
yOffset = cubeFaceHeight;
}
else if (za == -1)
{
//Back
xPixel = (int)((((xa + 1f) / 2f) - 1f) * cubeFaceWidth);
xOffset = 3 * cubeFaceWidth;
yPixel = (int)((((ya + 1f) / 2f)) * cubeFaceHeight);
yOffset = cubeFaceHeight;
}
else
{
Debug.LogWarning("Unknown face, something went wrong");
xPixel = 0;
yPixel = 0;
xOffset = 0;
yOffset = 0;
}
xPixel = Mathf.Abs(xPixel);
yPixel = Mathf.Abs(yPixel);
xPixel += xOffset;
yPixel += yOffset;
color = sourceTexture.GetPixel(xPixel, yPixel);
equiTexture.SetPixel(i, j, color);
}
}
equiTexture.Apply();
var bytes = equiTexture.EncodeToPNG();
Object.DestroyImmediate(equiTexture);
return bytes;
}
}
In order to utilize the GPU I created a shader that does the same conversion. It is much faster than running the conversion pixel by pixel on the CPU but unfortunately Unity imposes resolution limitations on cubemaps so it's usefulness is limited in scenarios when high resolution input image is to be used.
Shader "Conversion/CubemapToEquirectangular" {
Properties {
_MainTex ("Cubemap (RGB)", CUBE) = "" {}
}
Subshader {
Pass {
ZTest Always Cull Off ZWrite Off
Fog { Mode off }
CGPROGRAM
#pragma vertex vert
#pragma fragment frag
#pragma fragmentoption ARB_precision_hint_fastest
//#pragma fragmentoption ARB_precision_hint_nicest
#include "UnityCG.cginc"
#define PI 3.141592653589793
#define TWOPI 6.283185307179587
struct v2f {
float4 pos : POSITION;
float2 uv : TEXCOORD0;
};
samplerCUBE _MainTex;
v2f vert( appdata_img v )
{
v2f o;
o.pos = mul(UNITY_MATRIX_MVP, v.vertex);
o.uv = v.texcoord.xy * float2(TWOPI, PI);
return o;
}
fixed4 frag(v2f i) : COLOR
{
float theta = i.uv.y;
float phi = i.uv.x;
float3 unit = float3(0,0,0);
unit.x = sin(phi) * sin(theta) * -1;
unit.y = cos(theta) * -1;
unit.z = cos(phi) * sin(theta) * -1;
return texCUBE(_MainTex, unit);
}
ENDCG
}
}
Fallback Off
}
The quality of the resulting images can be greatly improved by either employing a more sophisticated method to estimate the color of a pixel during the conversion or by post processing the resulting image (or both, actually). For example an image of bigger size could be generated to apply a blur filter and then downsample it to the desired size.
I created a simple Unity project with two editor wizards that show how to properly utilize either the C# code or the shader shown above. Get it here:
https://github.com/Mapiarz/CubemapToEquirectangular
Remember to set proper import settings in Unity for your input images:
Point filtering
Truecolor format
Disable mipmaps
Non Power of 2: None (only for 2DTextures)
Enable Read/Write (only for 2DTextures)
cube2sphere automates the entire process. Example:
$ cube2sphere front.jpg back.jpg right.jpg left.jpg top.jpg bottom.jpg -r 2048 1024 -fTGA -ostitched
How can I save a frame via TangoService_connectOnFrameAvailable() and display it correctly on my computer? As this reference page mentions, the pixels are stored in the HAL_PIXEL_FORMAT_YV12 format. In my callback function for TangoService_connectOnFrameAvailable, I save the frame like this:
static void onColorFrameAvailable(void* context, TangoCameraId id, const TangoImageBuffer* buffer)
{
...
std::ofstream fp;
fp.open(imagefile, std::ios::out | std::ios::binary );
int offset = 0;
for(int i = 0; i < buffer->height*2 + 1; i++) {
fp.write((char*)(buffer->data + offset), buffer->width);
offset += buffer->stride;
}
fp.close();
}
Then to get rid of the meta data in the first row and to display the image I run:
$ dd if="input.raw" of="new.raw" bs=1 skip=1280
$ vooya new.raw
I was careful to make sure in vooya that the channel order is yvu. The resulting output is:
What am I doing wrong in saving the image and displaying it?
UPDATE per Mark Mullin's response:
int offset = buffer->stride; // header offset
// copy Y channel
for(int i = 0; i < buffer->height; i++) {
fp.write((char*)(buffer->data + offset), buffer->width);
offset += buffer->stride;
}
// copy V channel
for(int i = 0; i < buffer->height / 2; i++) {
fp.write((char*)(buffer->data + offset), buffer->width / 2);
offset += buffer->stride / 2;
}
// copy U channel
for(int i = 0; i < buffer->height / 2; i++) {
fp.write((char*)(buffer->data + offset), buffer->width / 2);
offset += buffer->stride / 2;
}
This now shows the picture below, but there are still some artifacts; I wonder if that's from the Tango tablet camera or my processing of the raw data... any thoughts?
Can't say exactly what you're doing wrong AND tango images often have artifacts in them - yours are new, but I often see baby blue as a color where glare seems to be annoying deeper systems, and as it begins to loose sync with the depth system under load, you'll often see what looks like a shiny grid (its the IR pattern, I think) - At the end, any rational attempt to handle the image with openCV etc failed, so I hand wrote the decoder with some help from SO thread here
That said, given imagebuffer contains a pointer to the raw data from Tango, and various other variables like height and stride are filled in from the data received in the callback, then this logic will create an RGBA map - yeah, I optimized the math in it, so it's a little ugly - it's slower but functionally equivalent twin is listed second. My own experience says its a horrible idea to try and do this decode right in the callback (I believe Tango is capable of loosing sync with the flash for depth for purely spiteful reasons), so mine runs at the render stage.
Fast
uchar* pData = TangoData::cameraImageBuffer;
uchar* iData = TangoData::cameraImageBufferRGBA;
int size = (int)(TangoData::imageBufferStride * TangoData::imageBufferHeight);
float invByte = 0.0039215686274509803921568627451; // ( 1 / 255)
int halfi, uvOffset, halfj, uvOffsetHalfj;
float y_scaled, v_scaled, u_scaled;
int uOffset = size / 4 + size;
int halfstride = TangoData::imageBufferStride / 2;
for (int i = 0; i < TangoData::imageBufferHeight; ++i)
{
halfi = i / 2;
uvOffset = halfi * halfstride;
for (int j = 0; j < TangoData::imageBufferWidth; ++j)
{
halfj = j / 2;
uvOffsetHalfj = uvOffset + halfj;
y_scaled = pData[i * TangoData::imageBufferStride + j] * invByte;
v_scaled = 2 * (pData[uvOffsetHalfj + size] * invByte - 0.5f) * Vmax;
u_scaled = 2 * (pData[uvOffsetHalfj + uOffset] * invByte - 0.5f) * Umax;
*iData++ = (uchar)((y_scaled + 1.13983f * v_scaled) * 255.0);;
*iData++ = (uchar)((y_scaled - 0.39465f * u_scaled - 0.58060f * v_scaled) * 255.0);
*iData++ = (uchar)((y_scaled + 2.03211f * u_scaled) * 255.0);
*iData++ = 255;
}
}
Understandable
for (int i = 0; i < TangoData::imageBufferHeight; ++i)
{
for (int j = 0; j < TangoData::imageBufferWidth; ++j)
{
uchar y = pData[i * image->stride + j];
uchar v = pData[(i / 2) * (TangoData::imageBufferStride / 2) + (j / 2) + size];
uchar u = pData[(i / 2) * (TangoData::imageBufferStride / 2) + (j / 2) + size + (size / 4)];
YUV2RGB(y, u, v);
*iData++ = y;
*iData++ = u;
*iData++ = v;
*iData++ = 255;
}
}
I think that there is a better way to do if you can to do it offline.
The best way to save the image should be something like this (don't forgot to create the folder Pictures or you won't save anything)
void onFrameAvailableRouter(void* context, TangoCameraId id, const TangoImageBuffer* buffer) {
//To write the image in a txt file.
std::stringstream name_stream;
name_stream.setf(std::ios_base::fixed, std::ios_base::floatfield);
name_stream.precision(3);
name_stream << "/storage/emulated/0/Pictures/"
<<cur_frame_timstamp_
<<".txt";
std::fstream f(name_stream.str().c_str(), std::ios::out | std::ios::binary);
// size = 1280*720*1.5 to save YUV or 1280*720 to save grayscale
int size = stride_ * height_ * 1.5;
f.write((const char *) buffer->data,size * sizeof(uint8_t));
f.close();
}
Then to convert the .txt file to png you can do this
inputFolder = "input"
outputFolderRGB = "output/rgb"
outputFolderGray = "output/gray"
input_filename = "timestamp.txt"
output_filename = "rgb.png"
allFile = listdir(inputFolder)
numberOfFile = len(allFile)
if "input" in glob.glob("*"):
if "output/rgb" in glob.glob("output/*"):
print ""
else:
makedirs("output/rgb")
if "output/gray" in glob.glob("output/*"):
print ""
else:
makedirs("output/gray")
#The output reportories are ready
for file in allFile:
count+=1
print "current file : ",count,"/",numberOfFile
input_filename = file
output_filename = input_filename[0:(len(input_filename)-3)]+"png"
# load file into buffer
data = np.fromfile(inputFolder+"/"+input_filename, dtype=np.uint8)
#To get RGB image
# create yuv image
yuv = np.ndarray((height + height / 2, width), dtype=np.uint8, buffer=data)
# create a height x width x channels matrix with the datatype uint8 for rgb image
img = np.zeros((height, width, channels), dtype=np.uint8);
# convert yuv image to rgb image
cv2.cvtColor(yuv, cv2.COLOR_YUV2BGRA_NV21, img, channels)
cv2.imwrite(outputFolderRGB+"/"+output_filename, img)
#If u saved the image in graysacale use this part instead
#yuvReal = np.ndarray((height, width), dtype=np.uint8, buffer=data)
#cv2.imwrite(outputFolderGray+"/"+output_filename, yuvReal)
else:
print "not any input"
You just have to put your .txt in a folder input
It's a python script but if you prefer a c++ version it's very close.
I am taking an image, loading it via the screen context, and changing it pixel by pixel. I have a number of different filters that I am applying to the images, but the last thing I need to do is shift the color balance (similar to Photoshop) to make the red more cyan.
The code below shows how I am taking the image, getting the data, and going through the r/g/b values pixel by pixel:
CGImageRef sourceImage = theImage.image.CGImage;
CFDataRef theData;
theData = CGDataProviderCopyData(CGImageGetDataProvider(sourceImage));
UInt8 *pixelData = (UInt8 *) CFDataGetBytePtr(theData);
int dataLength = CFDataGetLength(theData);
int red = 0;
int green = 1;
int blue = 2;
for (int index = 0; index < dataLength; index += 4) {
int r = pixelData[index + red];
int g = pixelData[index + green];
int b = pixelData[index + blue];
// the color balancing would go here...
if (r < 0) r = 0;
if (g < 0) g = 0;
if (b < 0) b = 0;
if (r > 255) r = 255;
if (g > 255) g = 255;
if (b > 255) b = 255;
pixelData[index + red] = r;
pixelData[index + green] = g;
pixelData[index + blue] = b;
}
CGContextRef context;
context = CGBitmapContextCreate(pixelData,
CGImageGetWidth(sourceImage),
CGImageGetHeight(sourceImage),
8,
CGImageGetBytesPerRow(sourceImage),
CGImageGetColorSpace(sourceImage),
kCGImageAlphaPremultipliedLast);
CGImageRef newCGImage = CGBitmapContextCreateImage(context);
UIImage *newImage = [UIImage imageWithCGImage:newCGImage];
CGContextRelease(context);
CFRelease(theData);
CGImageRelease(newCGImage);
theImage.image = newImage;
I am doing a number of other things to the pixel data (setting the levels, desaturating) but I need to shift the red value toward cyan.
In Photoshop, I am able to do it via:
Image: Adjustment: Color Balance
and setting it to -30 0 0
I have not been able to find an algorithm or an example explaining how this color shift is performed. I have tried to subtract 30 from every red value, or setting a hard maximum at 255 - 30 (225), but those seem to clip the colors and not shift the values... Right now, I'm just hacking around on it, but pointers to a reference would help out a lot.
(NB: I am not able to use an OpenGL solution because I have to take the same algorithm and convert it to PHP/gd for a web server version of this application)
What you need to do is subtract the red and then correct the "lightness" to match the old color, for whatever measure of lightness you choose. Something like this should work:
// First, calculate the current lightness.
float oldL = r * 0.30 + g * 0.59 + b * 0.11;
// Adjust the color components. This changes lightness.
r = r - 30;
if (r < 0) r = 0;
// Now correct the color back to the old lightness.
float newL = r * 0.30 + g * 0.59 + b * 0.11;
if (newL > 0) {
r = r * oldL / newL;
g = g * oldL / newL;
b = b * oldL / newL;
}
Note this behaves somewhat oddly when given pure red (it will leave it unchanged). OTOH, GIMP (version 2.6.11) does the same thing in its color balance tool so I'm not too worried about it.
I need to rotate an image by very small angle, like 1-5 degrees. Does OpenCV provide simple way of doing that? From reading docs i can assume that getAffineTransform() should be involved, but there is no direct example of doing something like:
IplImage *rotateImage( IplImage *source, double angle);
If you use OpenCV > 2.0 it is as easy as
using namespace cv;
Mat rotateImage(const Mat& source, double angle)
{
Point2f src_center(source.cols/2.0F, source.rows/2.0F);
Mat rot_mat = getRotationMatrix2D(src_center, angle, 1.0);
Mat dst;
warpAffine(source, dst, rot_mat, source.size());
return dst;
}
Note: angle is in degrees, not radians.
See the C++ interface documentation for more details and adapt as you need:
getRotationMatrix
warpAffine
Edit: To down voter: Please comment the reason for down voting a tried and tested code?
#include "cv.h"
#include "highgui.h"
#include "math.h"
int main( int argc, char** argv )
{
IplImage* src = cvLoadImage("lena.jpg", 1);
IplImage* dst = cvCloneImage( src );
int delta = 1;
int angle = 0;
int opt = 1; // 1： rotate & zoom
// 0: rotate only
double factor;
cvNamedWindow("src", 1);
cvShowImage("src", src);
for(;;)
{
float m[6];
CvMat M = cvMat(2, 3, CV_32F, m);
int w = src->width;
int h = src->height;
if(opt)
factor = (cos(angle*CV_PI/180.) + 1.05) * 2;
else
factor = 1;
m[0] = (float)(factor*cos(-angle*2*CV_PI/180.));
m[1] = (float)(factor*sin(-angle*2*CV_PI/180.));
m[3] = -m[1];
m[4] = m[0];
m[2] = w*0.5f;
m[5] = h*0.5f;
cvGetQuadrangleSubPix( src, dst, &M);
cvNamedWindow("dst", 1);
cvShowImage("dst", dst);
if( cvWaitKey(1) == 27 )
break;
angle =(int)(angle + delta) % 360;
}
return 0;
}
UPDATE: See the following code for rotation using warpaffine
https://code.google.com/p/opencvjp-sample/source/browse/trunk/cpp/affine2_cpp.cpp?r=48
#include <cv.h>
#include <highgui.h>
using namespace cv;
int
main(int argc, char **argv)
{
// (1)load a specified file as a 3-channel color image,
// set its ROI, and allocate a destination image
const string imagename = argc > 1 ? argv[1] : "../image/building.png";
Mat src_img = imread(imagename);
if(!src_img.data)
return -1;
Mat dst_img = src_img.clone();
// (2)set ROI
Rect roi_rect(cvRound(src_img.cols*0.25), cvRound(src_img.rows*0.25), cvRound(src_img.cols*0.5), cvRound(src_img.rows*0.5));
Mat src_roi(src_img, roi_rect);
Mat dst_roi(dst_img, roi_rect);
// (2)With specified three parameters (angle, rotation center, scale)
// calculate an affine transformation matrix by cv2DRotationMatrix
double angle = -45.0, scale = 1.0;
Point2d center(src_roi.cols*0.5, src_roi.rows*0.5);
const Mat affine_matrix = getRotationMatrix2D( center, angle, scale );
// (3)rotate the image by warpAffine taking the affine matrix
warpAffine(src_roi, dst_roi, affine_matrix, dst_roi.size(), INTER_LINEAR, BORDER_CONSTANT, Scalar::all(255));
// (4)show source and destination images with a rectangle indicating ROI
rectangle(src_img, roi_rect.tl(), roi_rect.br(), Scalar(255,0,255), 2);
namedWindow("src", CV_WINDOW_AUTOSIZE);
namedWindow("dst", CV_WINDOW_AUTOSIZE);
imshow("src", src_img);
imshow("dst", dst_img);
waitKey(0);
return 0;
}
Check my answer to a similar problem:
Rotating an image in C/C++
Essentially, use cvWarpAffine - I've described how to get the 2x3 transformation matrix from the angle in my previous answer.
Updating full answer for OpenCV 2.4 and up
// ROTATE p by R
/**
* Rotate p according to rotation matrix (from getRotationMatrix2D()) R
* #param R Rotation matrix from getRotationMatrix2D()
* #param p Point2f to rotate
* #return Returns rotated coordinates in a Point2f
*/
Point2f rotPoint(const Mat &R, const Point2f &p)
{
Point2f rp;
rp.x = (float)(R.at<double>(0,0)*p.x + R.at<double>(0,1)*p.y + R.at<double>(0,2));
rp.y = (float)(R.at<double>(1,0)*p.x + R.at<double>(1,1)*p.y + R.at<double>(1,2));
return rp;
}
//COMPUTE THE SIZE NEEDED TO LOSSLESSLY STORE A ROTATED IMAGE
/**
* Return the size needed to contain bounding box bb when rotated by R
* #param R Rotation matrix from getRotationMatrix2D()
* #param bb bounding box rectangle to be rotated by R
* #return Size of image(width,height) that will compleley contain bb when rotated by R
*/
Size rotatedImageBB(const Mat &R, const Rect &bb)
{
//Rotate the rectangle coordinates
vector<Point2f> rp;
rp.push_back(rotPoint(R,Point2f(bb.x,bb.y)));
rp.push_back(rotPoint(R,Point2f(bb.x + bb.width,bb.y)));
rp.push_back(rotPoint(R,Point2f(bb.x + bb.width,bb.y+bb.height)));
rp.push_back(rotPoint(R,Point2f(bb.x,bb.y+bb.height)));
//Find float bounding box r
float x = rp[0].x;
float y = rp[0].y;
float left = x, right = x, up = y, down = y;
for(int i = 1; i<4; ++i)
{
x = rp[i].x;
y = rp[i].y;
if(left > x) left = x;
if(right < x) right = x;
if(up > y) up = y;
if(down < y) down = y;
}
int w = (int)(right - left + 0.5);
int h = (int)(down - up + 0.5);
return Size(w,h);
}
/**
* Rotate region "fromroi" in image "fromI" a total of "angle" degrees and put it in "toI" if toI exists.
* If toI doesn't exist, create it such that it will hold the entire rotated region. Return toI, rotated imge
* This will put the rotated fromroi piece of fromI into the toI image
*
* #param fromI Input image to be rotated
* #param toI Output image if provided, (else if &toI = 0, it will create a Mat fill it with the rotated image roi, and return it).
* #param fromroi roi region in fromI to be rotated.
* #param angle Angle in degrees to rotate
* #return Rotated image (you can ignore if you passed in toI
*/
Mat rotateImage(const Mat &fromI, Mat *toI, const Rect &fromroi, double angle)
{
//CHECK STUFF
// you should protect against bad parameters here ... omitted ...
//MAKE OR GET THE "toI" MATRIX
Point2f cx((float)fromroi.x + (float)fromroi.width/2.0,fromroi.y +
(float)fromroi.height/2.0);
Mat R = getRotationMatrix2D(cx,angle,1);
Mat rotI;
if(toI)
rotI = *toI;
else
{
Size rs = rotatedImageBB(R, fromroi);
rotI.create(rs,fromI.type());
}
//ADJUST FOR SHIFTS
double wdiff = (double)((cx.x - rotI.cols/2.0));
double hdiff = (double)((cx.y - rotI.rows/2.0));
R.at<double>(0,2) -= wdiff; //Adjust the rotation point to the middle of the dst image
R.at<double>(1,2) -= hdiff;
//ROTATE
warpAffine(fromI, rotI, R, rotI.size(), INTER_CUBIC, BORDER_CONSTANT, Scalar::all(0));
//& OUT
return(rotI);
}
IplImage* rotate(double angle, float centreX, float centreY, IplImage* src, bool crop)
{
int w=src->width;
int h=src->height;
CvPoint2D32f centre;
centre.x = centreX;
centre.y = centreY;
CvMat* warp_mat = cvCreateMat(2, 3, CV_32FC1);
cv2DRotationMatrix(centre, angle, 1.0, warp_mat);
double m11= cvmGet(warp_mat,0,0);
double m12= cvmGet(warp_mat,0,1);
double m13= cvmGet(warp_mat,0,2);
double m21= cvmGet(warp_mat,1,0);
double m22= cvmGet(warp_mat,1,1);
double m23= cvmGet(warp_mat,1,2);
double m31= 0;
double m32= 0;
double m33= 1;
double x=0;
double y=0;
double u0= (m11*x + m12*y + m13)/(m31*x + m32*y + m33);
double v0= (m21*x + m22*y + m23)/(m31*x + m32*y + m33);
x=w;
y=0;
double u1= (m11*x + m12*y + m13)/(m31*x + m32*y + m33);
double v1= (m21*x + m22*y + m23)/(m31*x + m32*y + m33);
x=0;
y=h;
double u2= (m11*x + m12*y + m13)/(m31*x + m32*y + m33);
double v2= (m21*x + m22*y + m23)/(m31*x + m32*y + m33);
x=w;
y=h;
double u3= (m11*x + m12*y + m13)/(m31*x + m32*y + m33);
double v3= (m21*x + m22*y + m23)/(m31*x + m32*y + m33);
int left= MAX(MAX(u0,u2),0);
int right= MIN(MIN(u1,u3),w);
int top= MAX(MAX(v0,v1),0);
int bottom= MIN(MIN(v2,v3),h);
ASSERT(left<right&&top<bottom); // throw message?
if (left<right&&top<bottom)
{
IplImage* dst= cvCreateImage( cvGetSize(src), IPL_DEPTH_8U, src->nChannels);
cvWarpAffine(src, dst, warp_mat/*, CV_INTER_LINEAR + CV_WARP_FILL_OUTLIERS, cvScalarAll(0)*/);
if (crop) // crop and resize to initial size
{
IplImage* dst_crop= cvCreateImage(cvSize(right-left, bottom-top), IPL_DEPTH_8U, src->nChannels);
cvSetImageROI(dst,cvRect(left,top,right-left,bottom-top));
cvCopy(dst,dst_crop);
cvReleaseImage(&dst);
cvReleaseMat(&warp_mat);
//ver1
//return dst_crop;
// ver2 resize
IplImage* out= cvCreateImage(cvSize(w, h), IPL_DEPTH_8U, src->nChannels);
cvResize(dst_crop,out);
cvReleaseImage(&dst_crop);
return out;
}
else
{
/*cvLine( dst, cvPoint(left,top),cvPoint(left, bottom), cvScalar(0, 0, 255, 0) ,1,CV_AA);
cvLine( dst, cvPoint(right,top),cvPoint(right, bottom), cvScalar(0, 0, 255, 0) ,1,CV_AA);
cvLine( dst, cvPoint(left,top),cvPoint(right, top), cvScalar(0, 0, 255, 0) ,1,CV_AA);
cvLine( dst, cvPoint(left,bottom),cvPoint(right, bottom), cvScalar(0, 0, 255, 0) ,1,CV_AA);*/
cvReleaseMat(&warp_mat);
return dst;
}
}
else
{
return NULL; //assert?
}
}