lensfun  0.3.2.0
Public Member Functions | Static Public Member Functions | List of all members
lfModifier Struct Reference

A modifier object contains optimized data required to rectify a image. More...

#include <lensfun.h>

Public Member Functions

 lfModifier (const lfLens *lens, float crop, int width, int height)
 Create an empty image modifier object. More...
 
int Initialize (const lfLens *lens, lfPixelFormat format, float focal, float aperture, float distance, float scale, lfLensType targeom, int flags, bool reverse)
 Initialize the process of correcting aberrations in a image. More...
 
bool EnablePerspectiveCorrection (fvector x, fvector y, float d)
 Enable the perspective correction. More...
 
DEPRECATED void Destroy ()
 Destroy the modifier object. More...
 
void AddCoordCallback (lfModifyCoordFunc callback, int priority, void *data, size_t data_size)
 Add a user-defined callback to the coordinate correction chain. More...
 
void AddSubpixelCallback (lfSubpixelCoordFunc callback, int priority, void *data, size_t data_size)
 Add a user-defined callback to the subpixel coordinate rectification chain. More...
 
void AddColorCallback (lfModifyColorFunc callback, int priority, void *data, size_t data_size)
 Add a user-defined callback to the color modification chain. More...
 
bool AddSubpixelCallbackTCA (lfLensCalibTCA &model, bool reverse=false)
 Add the stock TCA correcting callback into the chain. More...
 
bool AddColorCallbackVignetting (lfLensCalibVignetting &model, lfPixelFormat format, bool reverse=false)
 Add the stock lens vignetting correcting callback into the chain. The vignetting correction callback always has a fixed priority of 250 when rectifying a image or 750 when doing reverse transform. More...
 
bool AddCoordCallbackDistortion (lfLensCalibDistortion &model, bool reverse=false)
 Add the stock lens distortion correcting callback into the chain. More...
 
bool AddCoordCallbackGeometry (lfLensType from, lfLensType to, float focal=0)
 Add the stock lens geometry rectification callback into the chain. More...
 
bool AddCoordCallbackScale (float scale, bool reverse=false)
 Add the stock image scaling callback into the chain. More...
 
float GetAutoScale (bool reverse)
 Compute the automatic scale factor for the image. More...
 
bool ApplyColorModification (void *pixels, float x, float y, int width, int height, int comp_role, int row_stride) const
 Image correction step 1: fix image colors. More...
 
bool ApplyGeometryDistortion (float xu, float yu, int width, int height, float *res) const
 Image correction step 2: apply the transforms on a block of pixel coordinates. More...
 
bool ApplySubpixelDistortion (float xu, float yu, int width, int height, float *res) const
 Image correction step 3: apply subpixel distortions. More...
 
bool ApplySubpixelGeometryDistortion (float xu, float yu, int width, int height, float *res) const
 Apply stage 2 & 3 in one step. More...
 

Static Public Member Functions

static DEPRECATED lfModifierCreate (const lfLens *lens, float crop, int width, int height)
 Create an empty image modifier object. More...
 

Detailed Description

A modifier object contains optimized data required to rectify a image.

You can either create an empty modifier object and then enable the required modification functions individually, or you can take a lens object, which contains a set of correction models, and create a modifier object from it.

Normally, you will create an instance of lfModifier and initialise immediately it with lfModifier::Initialize, passing a valid lens object in both cases. Users of the plain C interface will use lf_modifier_new() and lf_modifier_initialize() instead.

Every image modification has a corresponding inverse function, e.g. the library allows both to correct lens distortions and to simulate lens characteristics.

The normal order of applying a lens correction on a image is:

  1. Fix lens vignetting
  2. Fix chromatic aberrations (TCA)
  3. Fix lens distortion
  4. Fix lens geometry
  5. Fix perspective
  6. Scale the image

This is the theoretical order. But in reality, when using this library, the order is reversed (see How the corrections work for an explanation):

  1. Color
    1. Fix lens vignetting
  2. Coordinates
    1. Scale the image
    2. Fix perspective
    3. Fix lens geometry
    4. Fix lens distortion
  3. Subpixel coordinates
    1. Fix chromatic aberrations (TCA)

This process is divided into three stages.

In the first stage, the colors of the image pixels are fixed (vignetting). You pass a pointer to your pixel data, and it will be modified in place.

Then, the distortions introduced by the lens are removed (distortion and geometry), perspective is corrected if desired, and an additional scaling factor is applied if required. This operation requires building new image in a new allocated buffer: you cannot modify the image in place, or bad things will happen. Note that Lensfun does not provide pixel interpolation routines. You have to implement the lookup in the original image yourself.

And finally, in the subpixel distortion stage, application corrects transversal chromatic aberrations. For every target pixel coordinate the modifier will return you three new coordinates: the first will tell you the coordinates of the red component, second of the green, and third of the blue component. This again requires copying the image, but you can use the same buffers as in stage two, just in reverse order.

Of course, you can skip some stages of the process, e.g. if you, for example, don't want to change a fisheye image to a rectilinear you can omit that step.

Obviously, when simulating lens distortions, the modification stages must be applied in reverse order. While the library takes care to reverse the steps which are grouped into a single stage, the application must apply the stages themselves in reverse order.

HOWEVER. Doing it in three stages is not memory efficient, and is prone to error accumulation because you have to interpolate pixels twice - once during stage 2 and once during stage 3. To avoid this, it is sensful to do stages 2 & 3 in one step. In this case the output R,G,B coordinates from stage 2, which treats the colour channels equally, are fed directly into stage 3, which will correct the R,G,B coordinates further. ApplySubpixelGeometryDistortion() does this in a convenient fashion.

Examples:
lenstool.cpp.

Constructor & Destructor Documentation

lfModifier::lfModifier ( const lfLens lens,
float  crop,
int  width,
int  height 
)

Create an empty image modifier object.

Before using the returned object you must add the required modifier callbacks (see methods AddXXXCallback below).

You must provide the original image width/height even if you plan to correct just a part of the image.

Parameters
lensFor all modifications, the crop factor, aspect ratio, and center shift of this lens will be used.
cropThe crop factor for current camera. The distortion models will take this into account if lens models were measured on a camera with a different crop factor.
widthThe width of the image you want to correct.
heightThe height of the image you want to correct.

Member Function Documentation

void lfModifier::AddColorCallback ( lfModifyColorFunc  callback,
int  priority,
void *  data,
size_t  data_size 
)

Add a user-defined callback to the color modification chain.

Parameters
callbackThe callback to be called for every strip of pixels.
priorityCallback priority (0-999). Callbacks are always called in increasing priority order.
dataA pointer to additional user data. A copy of this data is made internally, so client application may do whatever it needs with this data after this call.
data_sizeUser data size in bytes. If data size is zero, the data is not copied and instead a verbatim copy of the 'data' parameter is passed to the callback.
bool lfModifier::AddColorCallbackVignetting ( lfLensCalibVignetting model,
lfPixelFormat  format,
bool  reverse = false 
)

Add the stock lens vignetting correcting callback into the chain. The vignetting correction callback always has a fixed priority of 250 when rectifying a image or 750 when doing reverse transform.

Parameters
modelLens vignetting model data.
formatPixel format of your image (bits per pixel component)
reverseIf true, the reverse model will be applied, e.g. simulate a lens' vignetting on a clean image.
Returns
True if vignetting model is valid and the callback was added to chain.
void lfModifier::AddCoordCallback ( lfModifyCoordFunc  callback,
int  priority,
void *  data,
size_t  data_size 
)

Add a user-defined callback to the coordinate correction chain.

Parameters
callbackThe callback to be called for every strip of pixels.
priorityCallback priority (0-999). Callbacks are always called in increasing priority order.
dataA pointer to additional user data. A copy of this data is made internally, so client application may do whatever it needs with this data after this call.
data_sizeUser data size in bytes. If data size is zero, the data is not copied and instead a verbatim copy of the 'data' parameter is passed to the callback.
bool lfModifier::AddCoordCallbackDistortion ( lfLensCalibDistortion model,
bool  reverse = false 
)

Add the stock lens distortion correcting callback into the chain.

The distortion correction callback always has a fixed priority of 750 when rectifying a image and 250 on reverse transform.

Parameters
modelLens distortion model data.
reverseIf true, the reverse model will be applied, e.g. simulate a lens' distortion on a clean image.
Returns
True if distortion model is valid and the callback was added to chain.
bool lfModifier::AddCoordCallbackGeometry ( lfLensType  from,
lfLensType  to,
float  focal = 0 
)

Add the stock lens geometry rectification callback into the chain.

The geometry correction callback always has a fixed priority of 500.

Parameters
fromThe lens model for source image.
toThe lens model for target image.
focalLens focal length in mm. Deprecated.
Returns
True if a library has a callback for given from->to conversion.
bool lfModifier::AddCoordCallbackScale ( float  scale,
bool  reverse = false 
)

Add the stock image scaling callback into the chain.

The scaling callback always has a fixed priority of 100. Note that scaling should be always the first operation to perform no matter if we're doing a forward or reverse transform.

Parameters
scaleImage scale. If equal to 0.0, the image is automatically scaled so that there won't be any unfilled gaps in the resulting image. Note that all coordinate distortion callbacks must be already added to the stack for this to work correctly!
reverseIf true, the reverse model will be applied.
Returns
True if the callback was added to chain.
void lfModifier::AddSubpixelCallback ( lfSubpixelCoordFunc  callback,
int  priority,
void *  data,
size_t  data_size 
)

Add a user-defined callback to the subpixel coordinate rectification chain.

Parameters
callbackThe callback to be called for every strip of pixels.
priorityCallback priority (0-999). Callbacks are always called in increasing priority order.
dataA pointer to additional user data. A copy of this data is made internally, so client application may do whatever it needs with this data after this call.
data_sizeUser data size in bytes. If data size is zero, the data is not copied and instead a verbatim copy of the 'data' parameter is passed to the callback.
bool lfModifier::AddSubpixelCallbackTCA ( lfLensCalibTCA model,
bool  reverse = false 
)

Add the stock TCA correcting callback into the chain.

The TCA correction callback always has a fixed priority of 500. The behaviour is undefined if you'll add more than one TCA correction callback to a modifier.

Parameters
modelLens TCA model data.
reverseIf true, the reverse model will be applied, e.g. simulate a lens' TCA on a clean image.
Returns
True if TCA model is valid and the callback was added to chain.
bool lfModifier::ApplyColorModification ( void *  pixels,
float  x,
float  y,
int  width,
int  height,
int  comp_role,
int  row_stride 
) const

Image correction step 1: fix image colors.

This currently is only vignetting transform.

Parameters
pixelsThis points to image pixels. The actual pixel format depends on both pixel_format and comp_role arguments. The results are stored in place in the same format. Warning: this array should be aligned at least on a 16-byte boundary.
xThe X coordinate of the corner of the block.
yThe Y coordinate of the corner of the block.
widthThe width of the image block in pixels.
heightThe height of the image block in pixels.
comp_roleThe role of every pixel component. This is a bitfield, made by one of the LF_CR_X macros which defines the roles of every pixel field. For example, LF_CR_4(RED,GREEN,BLUE,UNKNOWN) will define a RGBA (or RGBX) pixel format, and the UNKNOWN field will not be modified.
row_strideThe size of a image row in bytes. This can be actually different from width * pixel_width as some image formats use funny things like alignments, filler bytes etc.
Returns
true if return buffer has been altered, false if nothing to do
Examples:
lenstool.cpp.
bool lfModifier::ApplyGeometryDistortion ( float  xu,
float  yu,
int  width,
int  height,
float *  res 
) const

Image correction step 2: apply the transforms on a block of pixel coordinates.

The undistorted coordinates are computed for every pixel in a rectangular block: \((x_u, y_u), (x_u+1, y_u), \ldots, (x_u + \mathrm{width} - 1, y_u), (x_u, y_u + 1), \ldots, (x_u + \mathrm{width} - 1, y_u + \mathrm{height} - 1)\).

The corrected coordinates are put into the output buffer sequentially, X and Y values.

This routine has been designed to be safe to use in parallel from several threads.

Parameters
xuThe undistorted X coordinate of the start of the block of pixels.
yuThe undistorted Y coordinate of the start of the block of pixels.
widthThe width of the block in pixels.
heightThe height of the block in pixels.
resA pointer to an output array which receives the respective X and Y distorted coordinates for every pixel of the block. The size of this array must be at least width*height*2 elements. Warning: this array should be aligned at least on a 16-byte boundary.
Returns
true if return buffer has been filled, false if nothing to do
Examples:
lenstool.cpp.
bool lfModifier::ApplySubpixelDistortion ( float  xu,
float  yu,
int  width,
int  height,
float *  res 
) const

Image correction step 3: apply subpixel distortions.

The undistorted R,G,B coordinates are computed for every pixel in a square block: \((x_u, y_u), (x_u+1, y_u), \ldots, (x_u + \mathrm{width} - 1, y_u), (x_u, y_u + 1), \ldots, (x_u + \mathrm{width} - 1, y_u + \mathrm{height} - 1)\).

Returns the corrected coordinates separately for R/G/B channels The resulting coordinates are put into the output buffer sequentially, X and Y values.

This routine has been designed to be safe to use in parallel from several threads.

Parameters
xuThe undistorted X coordinate of the start of the block of pixels.
yuThe undistorted Y coordinate of the start of the block of pixels.
widthThe width of the block in pixels.
heightThe height of the block in pixels.
resA pointer to an output array which receives the respective X and Y distorted coordinates of the red, green and blue channels for every pixel of the block. The size of this array must be at least width*height*2*3 elements. Warning: this array should be aligned at least on a 16-byte boundary.
Returns
true if return buffer has been filled, false if nothing to do
Examples:
lenstool.cpp.
bool lfModifier::ApplySubpixelGeometryDistortion ( float  xu,
float  yu,
int  width,
int  height,
float *  res 
) const

Apply stage 2 & 3 in one step.

See the main comment to the lfModifier class. The undistorted (R, G, B) coordinates are computed for every pixel in a square block: \((x_u, y_u), (x_u+1, y_u), \ldots, (x_u + \mathrm{width} - 1, y_u), (x_u, y_u + 1), \ldots, (x_u + \mathrm{width} - 1, y_u + \mathrm{height} - 1)\).

Returns the corrected coordinates separately for R/G/B channels The resulting coordinates are put into the output buffer sequentially, X and Y values.

This routine has been designed to be safe to use in parallel from several threads.

Parameters
xuThe undistorted X coordinate of the start of the block of pixels.
yuThe undistorted Y coordinate of the start of the block of pixels.
widthThe width of the block in pixels.
heightThe height of the block in pixels.
resA pointer to an output array which receives the respective X and Y distorted coordinates for every pixel of the block. The size of this array must be at least width*height*2*3 elements. Warning: this array should be aligned at least on a 16-byte boundary.
Returns
true if return buffer has been filled, false if nothing to do
Examples:
lenstool.cpp.
static DEPRECATED lfModifier* lfModifier::Create ( const lfLens lens,
float  crop,
int  width,
int  height 
)
static

Create an empty image modifier object.

Before using the returned object you must add the required modifier callbacks (see methods AddXXXCallback below).

You must provide the original image width/height even if you plan to correct just a part of the image.

This function is deprecated and will be removed in the future. Please use the standard constructor instead.

Parameters
lensFor all modifications, the crop factor, aspect ratio, and center shift of this lens will be used.
cropThe crop factor for current camera. The distortion models will take this into account if lens models were measured on a camera with a different crop factor.
widthThe width of the image you want to correct.
heightThe height of the image you want to correct.
Returns
A new empty image modifier object.
DEPRECATED void lfModifier::Destroy ( )

Destroy the modifier object.

This function is deprecated and will be removed in the future. Please use the standard destructor instead.

bool lfModifier::EnablePerspectiveCorrection ( fvector  x,
fvector  y,
float  d 
)

Enable the perspective correction.

Depending on the number of control points given, there are three possible modes:

4 control points: c0 and c1 define one vertical lines, c2 and c3 the other. The focal length defined in the creation of the modifier is used to get the proper aspect ratio.

6 control points: c0 to c3 like above. c4 and c5 define a horizontal line. The focal length is used to get the proper aspect ratio.

8 control points: c0 to c5 like above. c6 and c7 define a second horizontal line. The focal length is not used.

5 control points: They all must lie on an ellipse that is actually a rotated circle. If they lie in a clockwise ordering, the vertex is assumed to be above the ellipse centre in the image. If they lie in a counter-clockwise ordering, the vertex is assumed to be below the ellipse. The focal length is needed to find the proper vertex.

7 control points: c0 to c4 like above. c5 and c6 define a horizontal line which is used to rotate the final image nicely.

If the lines constructed from the first four control points for the 4, 6 and 8 points case (or the last two for the 7 points case) are more horizontal than vertical, in all of the above, "horizontal" and "vertical" need to be swapped.

All control points must be given as pixel coordinates in the original image. For best precision, anti-distortion should have been applied before taking the coordinates. Moreover, fisheye images must have been transformed into rectilinear of the same focal length. In contrast, cropping, rotating, shifting, or scaling must be switched off.

Parameters
xThe x coordinates of the control points.
yThe y coordinates of the control points.
dThis parameter is supposed to be offered to the user as a slider. It can take values from -1 to +1. 0 denotes the perfect correction. -1 is the unchanged image. +1 is an increase of the tilting angle by 25%.
Returns
True if the perspective correction was enabled.
float lfModifier::GetAutoScale ( bool  reverse)

Compute the automatic scale factor for the image.

This expects that all coordinate distortion callbacks are already set up and working. The function will try at its best to find a scale value that will ensure that no pixels with coordinates out of range will get into the resulting image. But since this is an approximative method, the returned scale sometimes is a little less than the optimal scale (e.g. you can still get some black corners with some high-distortion cases).

Parameters
reverseIf true, the reverse scaling factor is computed.
int lfModifier::Initialize ( const lfLens lens,
lfPixelFormat  format,
float  focal,
float  aperture,
float  distance,
float  scale,
lfLensType  targeom,
int  flags,
bool  reverse 
)

Initialize the process of correcting aberrations in a image.

The modifier object will be set up to rectify all aberrations found in the lens description. Make sure the focal length, aperture, and focus distance are correct in order to ensure proper rectification.

Aperture and focus distance are only used for vignetting correction. Because the dependence of vignetting on focus distance is very small, and it is usually not available in EXIF data, you may give an approximative value here. If you really do not know, as a last resort, use "1000" as a default value.

Parameters
lensThe lens which aberrations you want to correct in a image. It should be the same lens object as the one passed to the lfModifier constructor.
formatPixel format of your image (bits per pixel component)
focalThe focal length in mm at which the image was taken.
apertureThe aperture (f-number) at which the image was taken.
distanceThe approximative focus distance in meters (distance > 0).
scaleAn additional scale factor to be applied onto the image (1.0 - no scaling; 0.0 - automatic scaling).
targeomTarget geometry. If LF_MODIFY_GEOMETRY is set in flags and targeom is different from lens->Type, a geometry conversion will be applied on the image.
flagsA set of flags (se LF_MODIFY_XXX) telling which distortions you want corrected. A value of LF_MODIFY_ALL orders correction of everything possible (will enable all correction models present in lens description).
reverseIf this parameter is true, a reverse transform will be prepared. That is, you take a undistorted image at input and convert it so that it will look as if it would be a shot made with lens.
Returns
A set of LF_MODIFY_XXX flags in effect. This is the flags argument with dropped bits for operations which are actually no-ops.
Examples:
lenstool.cpp.

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