Top |

## Functions

int | vips_conv () |

int | vips_convf () |

int | vips_convi () |

int | vips_conva () |

int | vips_convsep () |

int | vips_convasep () |

int | vips_compass () |

int | vips_gaussblur () |

int | vips_sharpen () |

int | vips_spcor () |

int | vips_fastcor () |

## Description

These operations convolve an image in some way, or are operations based on simple convolution, or are useful with convolution.

## Functions

### vips_conv ()

int vips_conv (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Optional arguments:

: VipsPrecision, calculation accuracy`precision`

:`layers`

`gint`

, number of layers for approximation:`cluster`

`gint`

, cluster lines closer than this distance

Convolution.

Perform a convolution of * in*
with

*. Each output pixel is calculated as:*

`mask`

1 |
sigma[i]{pixel[i] * mask[i]} / scale + offset |

where scale and offset are part of * mask*
.

By default, * precision*
is
VIPS_PRECISION_FLOAT. The output image
is always VIPS_FORMAT_FLOAT unless

*is VIPS_FORMAT_DOUBLE, in which case*

`in`

*is also VIPS_FORMAT_DOUBLE.*

`out`

If * precision*
is VIPS_PRECISION_INTEGER, then
elements of

*are converted to integers before convolution, using*

`mask`

`rint()`

,
and the output image
always has the same VipsBandFormat as the input image. For VIPS_FORMAT_UCHAR images and VIPS_PRECISION_INTEGER * precision*
,

`vips_conv()`

uses a fast vector path based on
fixed-point arithmetic. This can produce slightly different results.
Disable the vector path with `--vips-novector`

or `VIPS_NOVECTOR`

or
`vips_vector_set_enabled()`

.If * precision*
is VIPS_PRECISION_APPROXIMATE then, like
VIPS_PRECISION_INTEGER,

*is converted to int before convolution, and the output image always has the same VipsBandFormat as the input image.*

`mask`

Larger values for * layers*
give more accurate
results, but are slower. As

*approaches the mask radius, the accuracy will become close to exact convolution and the speed will drop to match. For many large masks, such as Gaussian,*

`layers`

*need be only 10% of this value and accuracy will still be good.*

`n_layers`

Smaller values of * cluster*
will give more accurate results, but be slower
and use more memory. 10% of the mask radius is a good rule of thumb.

See also: `vips_convsep()`

.

### vips_convf ()

int vips_convf (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Convolution. This is a low-level operation, see `vips_conv()`

for something
more convenient.

Perform a convolution of * in*
with

*. Each output pixel is calculated as sigma[i]{pixel[i] * mask[i]} / scale + offset, where scale and offset are part of*

`mask`

*.*

`mask`

The convolution is performed with floating-point arithmetic. The output image
is always VIPS_FORMAT_FLOAT unless * in*
is VIPS_FORMAT_DOUBLE, in which case

*is also VIPS_FORMAT_DOUBLE.*

`out`

See also: `vips_conv()`

.

### vips_convi ()

int vips_convi (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Integer convolution. This is a low-level operation, see `vips_conv()`

for
something more convenient.

* mask*
is converted to an integer mask with

`rint()`

of each element, rint of
scale and rint of offset. Each output pixel is then calculated as 1 |
sigma[i]{pixel[i] * mask[i]} / scale + offset |

The output image always has the same VipsBandFormat as the input image.

For VIPS_FORMAT_UCHAR images, `vips_convi()`

uses a fast vector path based on
half-float arithmetic. This can produce slightly different results.
Disable the vector path with `--vips-novector`

or `VIPS_NOVECTOR`

or
`vips_vector_set_enabled()`

.

See also: `vips_conv()`

.

### vips_conva ()

int vips_conva (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Optional arguments:

:`layers`

`gint`

, number of layers for approximation:`cluster`

`gint`

, cluster lines closer than this distance

Perform an approximate integer convolution of * in*
with

*. This is a low-level operation, see*

`mask`

`vips_conv()`

for something more convenient. The output image
always has the same VipsBandFormat as the input image.
Elements of * mask*
are converted to
integers before convolution.

Larger values for * layers*
give more accurate
results, but are slower. As

*approaches the mask radius, the accuracy will become close to exact convolution and the speed will drop to match. For many large masks, such as Gaussian,*

`layers`

*need be only 10% of this value and accuracy will still be good.*

`layers`

Smaller values of * cluster*
will give more accurate results, but be slower
and use more memory. 10% of the mask radius is a good rule of thumb.

See also: `vips_conv()`

.

### vips_convsep ()

int vips_convsep (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Optional arguments:

: calculation accuracy`precision`

: number of layers for approximation`layers`

: cluster lines closer than this distance`cluster`

Perform a separable convolution of * in*
with

*. See*

`mask`

`vips_conv()`

for a detailed description.The mask must be 1xn or nx1 elements.

The image is convolved twice: once with * mask*
and then again with

*rotated by 90 degrees. This is much faster for certain types of mask (gaussian blur, for example) than doing a full 2D convolution.*

`mask`

See also: `vips_conv()`

, `vips_gaussmat()`

.

### vips_convasep ()

int vips_convasep (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Optional arguments:

:`layers`

`gint`

, number of layers for approximation

Approximate separable integer convolution. This is a low-level operation, see
`vips_convsep()`

for something more convenient.

The image is convolved twice: once with * mask*
and then again with

*rotated by 90 degrees.*

`mask`

*must be 1xn or nx1 elements. Elements of*

`mask`

*are converted to integers before convolution.*

`mask`

Larger values for * layers*
give more accurate
results, but are slower. As

*approaches the mask radius, the accuracy will become close to exact convolution and the speed will drop to match. For many large masks, such as Gaussian,*

`layers`

*need be only 10% of this value and accuracy will still be good.*

`layers`

The output image always has the same VipsBandFormat as the input image.

See also: `vips_convsep()`

.

### vips_compass ()

int vips_compass (,`VipsImage *in`

,`VipsImage **out`

,`VipsImage *mask`

);`...`

Optional arguments:

:`times`

`gint`

, how many times to rotate and convolve: VipsAngle45, rotate mask by this much between colvolutions`angle`

: VipsCombine, combine results like this`combine`

: VipsPrecision, precision for blur, default float`precision`

:`layers`

`gint`

, number of layers for approximation:`cluster`

`gint`

, cluster lines closer than this distance

This convolves * in*
with

`mask`

*times, rotating*

`times`

*by*

`mask`

*each time. By default, it comvolves twice, rotating by 90 degrees, taking the maximum result.*

`angle`

See also: `vips_conv()`

.

### vips_gaussblur ()

int vips_gaussblur (,`VipsImage *in`

,`VipsImage **out`

,`double sigma`

);`...`

Optional arguments:

: VipsPrecision, precision for blur, default int`precision`

: minimum amplitude, default 0.2`min_ampl`

This operator runs `vips_gaussmat()`

and `vips_convsep()`

for you on an image.
Set * min_ampl*
smaller to generate a larger, more accurate mask. Set

*larger to make the blur more blurry.*

`sigma`

See also: `vips_gaussmat()`

, `vips_convsep()`

.

### vips_sharpen ()

int vips_sharpen (,`VipsImage *in`

,`VipsImage **out`

);`...`

Optional arguments:

: sigma of gaussian`sigma`

: flat/jaggy threshold`x1`

: maximum amount of brightening`y2`

: maximum amount of darkening`y3`

: slope for flat areas`m1`

: slope for jaggy areas`m2`

Selectively sharpen the L channel of a LAB image. The input image is transformed to VIPS_INTERPRETATION_LABS.

The operation performs a gaussian blur of radius * radius*
and subtracts
from

*to generate a high-frequency signal. This signal is passed through a lookup table formed from the five parameters and added back to*

`in`

*.*

`in`

The lookup table is formed like this:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 |
. ^ . y2 |- - - - - ----------- . | / . | / slope m2 . | .../ . -x1 | ... | . -------------------...----------------------> . | ... | x1 . |... slope m1 . / | . / m2 | . / | . / | . / | . / | . ______/ _ _ _ _ _ _ | -y3 . | |

For screen output, we suggest the following settings (the defaults):

1 2 3 4 5 6 |
sigma == 0.5 x1 == 2 y2 == 10 (don't brighten by more than 10 L*) y3 == 20 (can darken by up to 20 L*) m1 == 0 (no sharpening in flat areas) m2 == 3 (some sharpening in jaggy areas) |

If you want more or less sharpening, we suggest you just change the m2 parameter.

The * sigma*
parameter changes the width of the fringe and can be
adjusted according to the output printing resolution. As an approximate
guideline, use 0.5 for 4 pixels/mm (display resolution),
1.0 for 12 pixels/mm and 1.5 for 16 pixels/mm (300 dpi == 12
pixels/mm). These figures refer to the image raster, not the half-tone
resolution.

See also: `vips_conv()`

.

### vips_spcor ()

int vips_spcor (,`VipsImage *in`

,`VipsImage *ref`

,`VipsImage **out`

);`...`

Calculate a correlation surface.

* ref*
is placed at every position in

*and the correlation coefficient calculated. The output image is always float.*

`in`

The output image is the same size as the input. Extra input edge pixels are made by copying the existing edges outwards.

The correlation coefficient is calculated as:

1 2 3 4 |
sumij (ref(i,j)-mean(ref))(inkl(i,j)-mean(inkl)) c(k,l) = ------------------------------------------------ sqrt(sumij (ref(i,j)-mean(ref))^2) * sqrt(sumij (inkl(i,j)-mean(inkl))^2) |

where inkl is the area of * in*
centred at position (k,l).

from Niblack "An Introduction to Digital Image Processing", Prentice/Hall, pp 138.

If the number of bands differs, one of the images must have one band. In this case, an n-band image is formed from the one-band image by joining n copies of the one-band image together, and then the two n-band images are operated upon.

The output image is always float, unless either of the two inputs is double, in which case the output is also double.

See also: `vips_fastcor()`

.

### vips_fastcor ()

int vips_fastcor (,`VipsImage *in`

,`VipsImage *ref`

,`VipsImage **out`

);`...`

Calculate a fast correlation surface.

* ref*
is placed at every position in

*and the sum of squares of differences calculated.*

`in`

The output image is the same size as the input. Extra input edge pixels are made by copying the existing edges outwards.

If the number of bands differs, one of the images must have one band. In this case, an n-band image is formed from the one-band image by joining n copies of the one-band image together, and then the two n-band images are operated upon.

The output type is uint if both inputs are integer, float if both are float or complex, and double if either is double or double complex. In other words, the output type is just large enough to hold the whole range of possible values.

See also: `vips_spcor()`

.