Theia provides a basic Image<T> class to use for SfM and SLAM applications. The class is fairly lightweight, and is largely a wrapper for the CVD Image class but allows for more general use. It is the standard image class that used for feature detection within Theia.

class RGBImage
class GrayImage
class Image<T>

Images can be stored either as RGB or grayscale, denoted as RGBImage and GrayImage respectively. These are simply template specializations of the Image<T> class, so their interfaces will be nearly identical. Images can be easily read in from a file upon contruction or with the Read() command.

RGBImage my_rgb_image(img_filename);

// Or alternatively:
RGBImage my_rgb_image;

// ... do some stuff ... //

// Write the image back out.

Once an image is loaded, pixel values can be accessed with standard [] operators. Pixels are referenced in row then column order.

When accessing image pixels, either a RGBPixel or a float will be returned (for RBGImage and GrayImage respectively). The RGBPixel contains members red, green, and blue (all of type float).

// Load RGB and scale images.
RGBImage my_rgb_img("test_img.jpg");
GrayImage my_gray_img("test_img.jpg");

// Get the middle pixel location.
int middle_row = my_image.Rows()/2;
int middle_col = my_image.Cols()/2;

// Grab the middle pixel.
RGBPixel middle_rgb_pixel = my_rgb_img[middle_row][middle_col];
float middle_gray_pixel = my_gray_img[middle_row][middle_col];

// Output the RGB Pixel value.
LOG(INFO) << "red = " <<
          << " green = " <<
          << " blue = " <<;

// Output the grayscale pixel value.
LOG(INFO) << "gray = " << middle_gray_pixel;

You can also extra patches from images using the GetSubImage() function.

SubImage<T> GetSubImage(int row, int col, int num_rows, int num_cols)

Extracts a sub-image at the given row, col of the size (num_rows, num_cols). The row, col given should denote the top-left corner of the subwindow to extract. The extracted sub-image will be a reference to the patch in the original image, not a copy of it. However, the Image class that the patch was extracted from will still own the image so care must be taken when deleting images. GetSubImage() will return a SubImage<T> object of the same type that it is extracting from.

RGBImage my_img("...");

// Get the middle pixel location.
int middle_row = my_image.Rows()/2;
int middle_col = my_image.Cols()/2;

// Extract a 15x15 subimage centered at the middle pixel.
int patch_size = 15;
RGBSubImage middle_sub_img = my_img.SubImage(middle_row - patch_size/2,
                                             middle_col - patch_size/2,
class SubImage<T>
class RGBSubImage
class SubImage

We have also implemented some useful member functions of the Image class. For a full list of functions, theia/image/image.h

T* Data()
void Image<T>::GaussianBlur(double sigma, Image<T>* out_image) const
void Image<T>::GaussianBlur(double sigma)
Image<T> Image<T>::Integrate()
void Image<T>::Resize(int new_rows, int new_cols)
void Image<T>::Resize(double scale)
void Image<T>::HalfSample(Image<T>* out_image) const
Image<T> Image<T>::HalfSample() const
void Image<T>::TwoThirdsSample(Image<T>* out_image) const
Image<T> Image<T>::TwoThirdsSample() const


Previous topic


Next topic