stitcher.Stitcher Class Reference

Stitcher is responsible with stitching provided images. More...

Public Member Functions
None	__init__ (self, *, int target_image_mean=110, bool legacy_paste=False, float new_image_percentage=0.75, float min_homography_determinant=0.1, float filter_threshold=0.65, int max_iterations_before_fail=5, bool stitch_with_whole_image_on_fail=True, int initial_random_point_count=16, float ransac_threshold=0.5, initial_ransac_iterations=1250, ransac_steps=500)
	Creates a Stitcher object.
bool	stitch (self, np.ndarray image, bool whole_image=False)
	Stitches the given image to the full_image.

Public Attributes
ImageHistory	history
Optional[np.ndarray]	full_image
int	target_image_mean
bool	legacy_paste
float	new_image_percentage
float	min_homography_determinant
float	filter_threshold
bool	stitch_with_whole_image_on_fail
int	max_iterations_before_fail
int	initial_random_point_count
float	ransac_threshold
int	initial_ransac_iterations
int	ransac_steps
cv2.FlannBasedMatcher	flann_matcher
cv2.BFMatcher	bf_matcher
cv2.SIFT	sift_extractor
	history
	full_image
	ransac_threshold

Private Member Functions
np.ndarray	__preprocess_image (self, np.ndarray image)
	Preprocesses the given image.
Tuple[int, int]	__paste_image (self, np.ndarray image, int x_offset=0, int y_offset=0, bool whole_image=False)
	Pastes the image to the full_image.
KDTuple	__detect_and_compute (self, np.ndarray image)
	Detect and compute keypoints and descriptors of given image using SIFT.

Detailed Description

Stitcher is responsible with stitching provided images.

Stitching process goes like this:

• If the provided image is the first image, set self.image to that image and stop
• Otherwise, take the current image with last image and calculate the homography matrix between the images and applies it to the current image. Also calculates the where that image should be positioned in the final image.
• To paste the image, stitcher first creates a new canvas and pastes the old image and the warped image to the correct position. There are two blending modes: legacy mode and the new mode. Legacy mode is just alpha blending. The new mode however, mixes the overlapping parts of two images by a predefined constant.

Constructor & Destructor Documentation

__init__()

None stitcher.Stitcher.__init__	(		self,
		*	,
		int	target_image_mean = 110,
		bool	legacy_paste = False,
		float	new_image_percentage = 0.75,
		float	min_homography_determinant = 0.1,
		float	filter_threshold = 0.65,
		int	max_iterations_before_fail = 5,
		bool	stitch_with_whole_image_on_fail = True,
		int	initial_random_point_count = 16,
		float	ransac_threshold = 0.5,
			initial_ransac_iterations = 1250,
			ransac_steps = 500 )

Creates a Stitcher object.

Parameters

target_image_mean	The target brightness value of input images
legacy_paste	Turn on legacy paste mode. The difference between legacy paste and new paste modes are explained in the class description.
new_image_percentage	Used when mixing two images in the new paste mode. Does a linear interpolation between the pixels of the old image and the new image. Essentially c*new + (1-c)*old where c is this constant.
min_homography_determinant	Minimum allowed value for the determinant of the homography matrix.
filter_threshold	A threshold value used to filter in the Util.match function.
max_iterations_before_fail	If a step fails, instead of erroring out, the stitcher tries stitching again with different values until a match is found. This is the number of tries before failure.
stitch_with_whole_image_on_fail	If stitcher fails to stitch and image with its predecessor image and this option is enabled, it tries to stitch it with the whole image and starts the whole process again.
initial_random_point_count	The ransac works by picking some random points and checking if they are "good". On each fail the stitcher decreases this value by However, the minimum point count is 4.
ransac_threshold	A threshold value to determine if a point is "good" in the ransac algorithm
initial_ransac_iterations	The initial iteration count for ransac
ransac_steps	After each failure, the iteration count increases by this value

Member Function Documentation

__detect_and_compute()

KDTuple stitcher.Stitcher.__detect_and_compute ( self, np.ndarray image )

private

Detect and compute keypoints and descriptors of given image using SIFT.

Parameters

image

Image to find keypoints and descriptors

Returns

Keypoints and descriptors as a KDTuple object

__paste_image()

Tuple[int, int] stitcher.Stitcher.__paste_image ( self, np.ndarray image, int x_offset = 0, int y_offset = 0, bool whole_image = False )

private

Pastes the image to the full_image.

Parameters

image	Image to be pasted
x_offset	Horizontal offset from the last pasted image
y_offset	Vertical offset from the last pasted image
whole_image	If set to true, the offset applies to the whole image, not to the last pasted image

Returns

The total offset of the image as a tuple of ints

__preprocess_image()

np.ndarray stitcher.Stitcher.__preprocess_image ( self, np.ndarray image )

private

Preprocesses the given image.

Converts it from BGR to BGRA and applies a mean shift to shift the mean of the image to the target_brightness parameter

Parameters

image

Input image

Returns

Preprocessed image

stitch()

bool stitcher.Stitcher.stitch	(		self,
		np.ndarray	image,
		bool	whole_image = False )

Stitches the given image to the full_image.

   The process is explained in the class description.

Parameters

image	Image to stitch
whole_image	If set to true, it uses full_image instead of the last stitched image for stitching

Returns

True if stitching was successful, False otherwise

---

The documentation for this class was generated from the following file:

• src/stitcher.py