stim/image/image.h

#ifndef STIM_IMAGE_H
#define STIM_IMAGE_H
#ifdef USING_OPENCV
	//#include <opencv2/core/core.hpp>
	//#include <opencv2/highgui/highgui.hpp>
	#include <opencv2/opencv.hpp>
#endif
#include <vector>
#include <iostream>
#include <limits>
#include <typeinfo>
#include <fstream>
#include <cstring>
namespace stim{
/// This static class provides the STIM interface for loading, saving, and storing 2D images.
/// Data is stored in an interleaved (BIP) format (default for saving and loading is RGB).
//currently this interface uses CImg
//	T = data type (usually unsigned char)
template <class T>
class image{
	T* img;										//pointer to the image data (interleaved RGB for color)
	size_t R[3];
	inline size_t X() const { return R[1]; }
	inline size_t Y() const { return R[2]; }
	inline size_t C() const { return R[0]; }
	void init(){								//initializes all variables, assumes no memory is allocated
		memset(R, 0, sizeof(size_t) * 3);		//set the resolution and number of channels to zero
		img = NULL;
	}
	void unalloc(){								//frees any resources associated with the image
		if(img)	free(img);						//if memory has been allocated, free it
	}
	void clear(){								//clears all image data
		unalloc();								//unallocate previous memory
		init();									//re-initialize the variables
	}
	void allocate(){
		unalloc();
		img = (T*) malloc( sizeof(T) * R[0] * R[1] * R[2] );	//allocate memory
	}
	void allocate(size_t x, size_t y, size_t c){	//allocate memory based on the resolution
		R[0] = c; R[1] = x; R[2] = y;				//set the resolution
		allocate();									//allocate memory
	}
	inline size_t idx(size_t x, size_t y, size_t c = 0) const {
		return y * R[0] * R[1] + x * R[0] + c;
	}
#ifdef USING_OPENCV
	int cv_type(){
		if(typeid(T) == typeid(unsigned char))		return CV_MAKETYPE(CV_8U, (int)C());
		if(typeid(T) == typeid(char))				return CV_MAKETYPE(CV_8S, (int)C());
		if(typeid(T) == typeid(unsigned short))		return CV_MAKETYPE(CV_16U, (int)C());
		if(typeid(T) == typeid(short))				return CV_MAKETYPE(CV_16S, (int)C());
		if(typeid(T) == typeid(int))				return CV_MAKETYPE(CV_32S, (int)C());
		if(typeid(T) == typeid(float))				return CV_MAKETYPE(CV_32F, (int)C());
		if(typeid(T) == typeid(double))				return CV_MAKETYPE(CV_64F, (int)C());
		std::cout<<"ERROR in stim::image::cv_type - no valid data type found"<<std::endl;
		exit(1);
	}
#endif
	/// Returns the value for "white" based on the dynamic range (assumes white is 1.0 for floating point images)
	T white(){
		return std::numeric_limits<T>::max();
	}
public:
	size_t bytes() { return size() * sizeof(T); }
	/// Default constructor - creates an empty image object
	image(){ init(); }							//initialize all variables to zero, don't allocate any memory
	/// Constructor with a filename - loads the specified file
	image(std::string filename){				//constructor initialize the image with an image file
		init();
		load(filename);
	}
	/// Create a new image from scratch given a number of samples and channels
	image(size_t x, size_t y = 1, size_t c = 1){
		init();
		allocate(x, y, c);
	}
	/// Create a new image with the data given in 'data'
	image(T* data, size_t x, size_t y, size_t c = 1){
		init();
		allocate(x, y, c);
		memcpy(img, data, bytes());
	}
	/// Copy constructor - duplicates an image object
	image(const stim::image<T>& I){
		init();
		allocate(I.X(), I.Y(), I.C());
		memcpy(img, I.img, bytes());
	}
	/// Destructor - clear memory
	~image(){
		free(img);
	}
	///Resize an image - this function looks like it hasn't been implemented
	void resize(size_t x, size_t y, size_t c = 1) {
		allocate(x, y, c);
	}
	stim::image<T>& operator=(const stim::image<T>& I){
		init();
		if(&I == this)									//handle self-assignment
			return *this;
		allocate(I.X(), I.Y(), I.C());
		memcpy(img, I.img, bytes());
		return *this;
	}
	//save a Netpbm file
	void load_netpbm(std::string filename) {
		std::ifstream infile(filename.c_str(), std::ios::in | std::ios::binary);		//open an output file
		if (!infile) {
			std::cout << "Error opening input file in image::load_netpbm()" << std::endl;
			exit(1);
		}
		size_t nc;													//allocate space for the number of channels
		char format[2];												//allocate space to hold the image format tag
		infile.read(format, 2);										//read the image format tag
		infile.seekg(1, std::ios::cur);								//skip the newline character
		if (format[0] != 'P') {
			std::cout << "Error in image::load_netpbm() - file format tag is invalid: " << format[0] << format[1] << std::endl;
			exit(1);
		}
		if (format[1] == '5') nc = 1;								//get the number of channels from the format flag
		else if (format[1] == '6') nc = 3;
		else {
			std::cout << "Error in image::load_netpbm() - file format tag is invalid: " << format[0] << format[1] << std::endl;
			exit(1);
		}
		unsigned char c;								//stores a character
		while (infile.peek() == '#') {					//if the next character indicates the start of a comment
			while (true) {
				c = infile.get();
				if (c == 0x0A) break;
			}
		}
		std::string sw;									//create a string to store the width of the image
		while(true){
			c = infile.get();							//get a single character
			if (c == ' ') break;						//exit if we've encountered a space
			sw.push_back(c);							//push the character on to the string
		}
		size_t w = atoi(sw.c_str());					//convert the string into an integer
		std::string sh;
		while (true) {
			c = infile.get();
			if (c == 0x0A) break;
			sh.push_back(c);
		}
		while (true) {									//skip the maximum value
			c = infile.get();
			if (c == 0x0A) break;
		}
		size_t h = atoi(sh.c_str());					//convert the string into an integer
		allocate(w, h, nc);													//allocate space for the image
		unsigned char* buffer = (unsigned char*)malloc(w * h * nc);			//create a buffer to store the read data
		infile.read((char*)buffer, size());									//copy the binary data from the file to the image
		infile.close();														//close the file
		for (size_t n = 0; n < size(); n++) img[n] = (T)buffer[n];			//copy the buffer data into the image
		free(buffer);														//free the buffer array
	}
	
#ifdef USING_OPENCV
	void from_opencv(unsigned char* buffer, size_t width, size_t height) {
		allocate(width, height, 3);
		T value;
		size_t i;
		for (size_t c = 0; c < C(); c++) {								//copy directly
			for (size_t y = 0; y < Y(); y++) {
				for (size_t x = 0; x < X(); x++) {
					i = y * X() * C() + x * C() + (2 - c);
					value = buffer[i];
					img[idx(x, y, c)] = value;
				}
			}
		}
	}
#endif
	/// Load an image from a file
	void load(std::string filename){
#ifdef USING_OPENCV
		cv::Mat cvImage = cv::imread(filename, CV_LOAD_IMAGE_UNCHANGED);	//use OpenCV to open the image file
		if(!cvImage.data){
			std::cout<<"ERROR stim::image::load() - unable to find image "<<filename<<std::endl;
			exit(1);
		}
		int cols = cvImage.cols;
		int rows = cvImage.rows;
		int channels = cvImage.channels();
		allocate(cols, rows, channels);			//allocate space for the image
		unsigned char* cv_ptr = (unsigned char*)cvImage.data;
		if(C() == 1)														//if this is a single-color image, just copy the data
			memcpy(img, cv_ptr, bytes());
		if(C() == 3)														//if this is a 3-color image, OpenCV uses BGR interleaving
			from_opencv(cv_ptr, X(), Y());
#else
		load_netpbm(filename);
#endif
	}
	//save a Netpbm file
	void save_netpbm(std::string filename) {
		std::ofstream outfile(filename.c_str(), std::ios::out | std::ios::binary);		//open an output file
		if(!outfile) {
			std::cout << "Error generating output file in image::save_netpbm()" << std::endl;
			exit(1);
		}
		if (sizeof(T) != 1) {
			std::cout << "Error in image::save_netpbm() - data type must be 8-bit integer." << std::endl;
			exit(1);
		}
		std::string format;
		if (channels() == 1) outfile << "P5" << (char)0x0A;			//output P5 if the file is grayscale
		else if (channels() == 3) outfile << "P6" << (char)0x0A;		//output P6 if the file is color
		else {
			std::cout << "Error in image::save_netpbm() - data must be grayscale or RGB." << std::endl;
			exit(1);
		}
		size_t w = width();
		size_t h = height();
		outfile << w << " " << h << (char)0x0A;			//save the width and height
		outfile << "255" << (char)0x0A;								//output the maximum value
		outfile.write((const char*)img, size());			//write the binary data
		outfile.close();
	}
	//save a file
	void save(std::string filename){
#ifdef USING_OPENCV
		//OpenCV uses an interleaved format, so convert first and then output
		T* buffer = (T*) malloc(bytes());
		if(C() == 1)
			memcpy(buffer, img, bytes());
		else if(C() == 3)
			get_interleaved_bgr(buffer);
		cv::Mat cvImage((int)Y(), (int)X(), cv_type(), buffer);
		cv::imwrite(filename, cvImage);
		free(buffer);
#else
		save_netpbm(filename);
#endif
	}
	void set_interleaved(T* buffer, size_t width, size_t height, size_t channels){
		allocate(width, height, channels);
		memcpy(img, buffer, bytes());
	}
	//create an image from an interleaved buffer
	void set_interleaved_rgb(T* buffer, size_t width, size_t height){
		set_interleaved(buffer, width, height, 3);
	}
	void set_interleaved_bgr(T* buffer, size_t width, size_t height){
		allocate(width, height, 3);
		T value;
		size_t i;
		for(size_t c = 0; c < C(); c++){								//copy directly
			for(size_t y = 0; y < Y(); y++){
				for(size_t x = 0; x < X(); x++){
					i = y * X() * C() + x * C() + (2-c);
					value = buffer[i];
					img[idx(x, y, c)] = value;
				}
			}
		}
	}
	void set_interleaved(T* buffer, size_t width, size_t height){
		set_interleaved_rgb(buffer, width, height);
	}
	void get_interleaved_bgr(T* data){
		//for each channel
		for(size_t y = 0; y < Y(); y++){
			for(size_t x = 0; x < X(); x++){
				for(size_t c = 0; c < C(); c++){
					data[y * X() * C() + x * C() + (2-c)] = img[idx(x, y, c)];
				}
			}
		}
	}
	void get_interleaved_rgb(T* data){
		memcpy(data, img, bytes());
	}
	//copies data in the given channel order as a non-interleaved image
	void get_noninterleaved(T* data){
		//for each channel
		for(size_t y = 0; y < Y(); y++){
			for(size_t x = 0; x < X(); x++){
				for(size_t c = 0; c < C(); c++){
					data[c * Y() * X() + y * X() + x] = img[idx(x, y, c)];
				}
			}
		}
	}
	/// Return an image representing a specified channel
	/// @param c is the channel to be returned
	image<T> channel(size_t c) const {		
		image<T> r(X(), Y(), 1);				//create a new image
		for(size_t x = 0; x < X(); x++){
			for(size_t y = 0; y < Y(); y++){
				r.img[r.idx(x, y, 0)] = img[idx(x, y, c)];
			}
		}
		return r;
	}
	/// Returns an std::vector containing each channel as a separate image
	std::vector< image<T> > split() const {
		std::vector< image<T> > r;			//create an image array
		r.resize(C());						//create images for each channel
		for (size_t c = 0; c < C(); c++) {	//for each channel
			r[c] = channel(c);				//copy the channel image to the array
		}
		return r;
	}
	/// Merge a series of single-channel images into a multi-channel image
	void merge(std::vector< image<T> >& list) {
		size_t x = list[0].width();				//calculate the size of the image
		size_t y = list[0].height();
		allocate(x, y, list.size());			//re-allocate the image
		for (size_t c = 0; c < list.size(); c++)		//for each channel
			set_channel(list[c].channel(0).data(), c);	//insert the channel into the output image
	}
	T& operator()(size_t x, size_t y, size_t c = 0){
		return img[idx(x, y, c)];
	}
	/// Set all elements in the image to a given scalar value
	/// @param v is the value used to set all values in the image
	image<T> operator=(T v){
		size_t N = size();
		for(size_t n = 0; n < N; n++)
			img[n] = v;
		return *this;
	}
	/// invert the image, given a specified maximum value (ex. maxval = 255, I' = 255 - I)
	/*image<T> invert(T maxval) {
		image<T> result(width(), height(), channels());		//create a new image
		size_t N = size();									//get the number of elements in the image
		for (size_t n = 0; n < N; n++)
			result.data()[n] = maxval - img[n];				//perform the inversion and save the result to the new image
		return result;
	}*/
	/// Stretch the contrast of the image such that the minimum and maximum intensity match the given values
	image<T> stretch(T low, T high) {
		T maxval = maxv();
		T minval = minv();
		
		image<T> result = *this;				//create a new image for output
		size_t N = size();						//get the number of values in the image
		double range = maxval - minval;			//calculate the current range of the image
		double desired_range = high - low;		//calculate the desired range of the image
		for (size_t n = 0; n < N; n++) {		//for each element in the image
			result.data()[n] = desired_range * (img[n] - minval) / range + low;
		}
		return result;
	}
	/// Add a border of width w with the given value around the image
	/// @param w specifies the total size of the border
	/// @param T is the pixel value (all channels will be the same)
	image<T> border(size_t w, T value = 0) {
		image<T> result(width() + w * 2, height() + w * 2, channels());						//create an output image
		result = value;														//assign the border value to all pixels in the new image
		for (size_t y = 0; y < height(); y++) {								//for each pixel in the original image
			for (size_t x = 0; x < width(); x++) {
				size_t n = (y + w) * (width() + w * 2) + x + w;				//calculate the index of the corresponding pixel in the result image
				size_t n0 = idx(x,y);										//calculate the index for this pixel in the original image
				result.data()[n] = img[n0];									// copy the original image to the result image afer the border area
			}
		}
		return result;
	}
	/// Copy the given data to the specified channel
	/// @param c is the channel number that the data will be copied to
	/// @param buffer is a pointer to the image to be copied to channel c
	void set_channel(T* buffer, size_t c){
		size_t x, y;
		for(y = 0; y < Y(); y++){
			for(x = 0; x < X(); x++){
				img[idx(x, y, c)] = buffer[y * X() + x];
			}
		}
	}
	size_t channels() const{
		return C();
	}
	size_t width() const{
		return X();
	}
	size_t height() const{
		return Y();
	}
	T* data(){
		return img;
	}
	//returns the size (number of values) of the image
	size_t size(){ return C() * X() * Y(); }
	/// Returns the number of nonzero values
	size_t nnz(){
		size_t N = X() * Y() * C();
		size_t nz = 0;
		for(size_t n = 0; n < N; n++)
			if(img[n] != 0) nz++;
		return nz;	//return the number of nonzero pixels
	}
	//this function returns indices of pixels that have nonzero values
	std::vector<size_t> sparse_idx(){
		std::vector<size_t> s;				//allocate an array
		s.resize(nnz());					//allocate space in the array
		size_t N = size();
		//size_t C = channels();
		//T* ptr = img.data();				//get a pointer to the image data
		size_t i = 0;
		for(size_t n = 0; n < N; n++){
			if(img[n] != 0){
				s[i] = n;
				i++;
			}
		}
		return s;			//return the index list
	}
	/// Returns the maximum pixel value in the image
	T maxv(){
		T max_val = img[0];				//initialize the maximum value to the first one
		size_t N = size();	//get the number of pixels
		for (size_t n=0; n<N; n++){		//for every value
			if (img[n] > max_val){			//if the value is higher than the current max
				max_val = img[n];
			}
		}
		return max_val;
	}
	/// Returns the maximum pixel value in the image
	T minv(){
		T min_val = img[0];				//initialize the maximum value to the first one
		size_t N = size();	//get the number of pixels
		for (size_t n=0; n<N; n++){		//for every value
			if (img[n] < min_val){			//if the value is higher than the current max
				min_val = img[n];
			}
		}
		return min_val;
	}
	/// Invert an image by calculating I1 = alpha - I0, where alpha is the maximum image value
	image<T> invert(T white_val){
		size_t N = size();						//calculate the total number of values in the image
		image<T> r(X(), Y(), C());				//allocate space for the resulting image
		for(size_t n = 0; n < N; n++)
			r.img[n] = white_val - img[n];		//perform the inversion
		return r;								//return the inverted image
	}
	image<T> crop(size_t x0, size_t y0, size_t w, size_t h){
		image<T> result(w, h, C());								//create the output cropped image
		size_t srci;
		size_t dsti;
		size_t line_bytes = w * C();							//calculate the number of bytes in a line
		for (size_t yi = 0; yi < h; yi++) {						//for each row in the cropped image
			srci = (y0 + yi) * X() * C() + x0 * C();			//calculate the source index
			dsti = yi * w * C();								//calculate the destination index
			memcpy(&result.img[dsti], &img[srci], line_bytes);	//copy the data
		}
		return result;
	}
	//crop regions given by an array of 1D index values
	std::vector<image<T>> crop_idx(size_t w, size_t h, std::vector<size_t> idx) {
		std::vector<image<T>> result(idx.size());										//create an array of image files to return
		for (size_t i = 0; i < idx.size(); i++) {										//for each specified index point
			size_t y = idx[i] / X();													//calculate the y coordinate from the 1D index (center of ROI)
			size_t x = idx[i] - y * X();												//calculate the x coordinate (center of ROI)
			y -= w / 2;																	//update x and y values to reflect the lower corner of the ROI
			x -= h / 2;
			result[i] = crop(x, y, w, h);												//get the cropped image and store it in the result array
		}
		return result;
	}
	image<T> srgb2lab(){
		std::cout<<"ERROR stim::image::srgb2lab - function has been broken, re-implement."<<std::endl;
		exit(1);
	}
	image<T> convolve2(image<T> mask){
		std::cout<<"ERROR stim::image::convolve2 - function has been broken, and shouldn't really be in here."<<std::endl;
		exit(1);
	}
	image<T> rotate(float angle, float cx, float cy){
		std::cout<<"ERROR stim::image::rotate - function has been broken, and shouldn't really be in here."<<std::endl;
		exit(1);
	}
	// leila's code for non_interleaving data in 3D
	//create an data set from an interleaved buffer
	void set_interleaved3(T* buffer, size_t width, size_t height, size_t depth, size_t channels = 3){
		std::cout<<"ERROR stim::image::set_interleaved3 - stim::image no longer supports 3D images."<<std::endl;
		exit(1);
	}
	/// Casting operator, casts every value in an image to a different data type V
	template<typename V>
	operator image<V>() {
		image<V> r(X(), Y(), C());					//create a new image
		std::copy(img, img + size(), r.data());		//copy and cast the data
		return r;									//return the new image
	}
};
};		//end namespace stim
#endif