median_imagemedian_imageMedianImageMedianImagemedian_image (Operator)


median_imagemedian_imageMedianImageMedianImagemedian_image — Compute a median filter with various masks.


median_image(Image : ImageMedian : MaskType, Radius, Margin : )

Herror median_image(const Hobject Image, Hobject* ImageMedian, const char* MaskType, const Hlong Radius, const char* Margin)

Herror T_median_image(const Hobject Image, Hobject* ImageMedian, const Htuple MaskType, const Htuple Radius, const Htuple Margin)

void MedianImage(const HObject& Image, HObject* ImageMedian, const HTuple& MaskType, const HTuple& Radius, const HTuple& Margin)

HImage HImage::MedianImage(const HString& MaskType, Hlong Radius, const HTuple& Margin) const

HImage HImage::MedianImage(const HString& MaskType, Hlong Radius, const HString& Margin) const

HImage HImage::MedianImage(const char* MaskType, Hlong Radius, const char* Margin) const

HImage HImage::MedianImage(const wchar_t* MaskType, Hlong Radius, const wchar_t* Margin) const   (Windows only)

static void HOperatorSet.MedianImage(HObject image, out HObject imageMedian, HTuple maskType, HTuple radius, HTuple margin)

HImage HImage.MedianImage(string maskType, int radius, HTuple margin)

HImage HImage.MedianImage(string maskType, int radius, string margin)

def median_image(image: HObject, mask_type: str, radius: int, margin: Union[int, float, str]) -> HObject


median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image performs a median filter on the input image ImageImageImageImageimageimage with a square or circular mask and returns the filtered image in ImageMedianImageMedianImageMedianImageMedianimageMedianimage_median. The shape of the mask can be selected with MaskTypeMaskTypeMaskTypeMaskTypemaskTypemask_type. The radius of the mask can be selected with RadiusRadiusRadiusRadiusradiusradius.

Conceptually, the median filter sorts all gray values within the mask in ascending order and then selects the median of the gray values. The median is the “middle” one of the sorted gray values, i.e., the gray value with rank (position) (N - 1) / 2 + 1 of the sorted gray values, where N denotes the number of pixels covered by the filter mask. Here, the rank 1 corresponds to the smallest gray value and the rank N corresponds to the largest gray value within the mask (see also rank_imagerank_imageRankImageRankImageRankImagerank_image).

The filter mask is determined by RadiusRadiusRadiusRadiusradiusradius, defining the size, and MaskTypeMaskTypeMaskTypeMaskTypemaskTypemask_type, defining the shape of the mask. For latter one, the following options are available:

median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image can be used, for example, to smooth images, to suppress unwanted objects (e.g., point-like or line-like structures) that are smaller than the mask, and can therefore be used to estimate the background illumination for a shading correction or as a preprocessing step for the dynamic threshold operation (see dyn_thresholddyn_thresholdDynThresholdDynThresholdDynThresholddyn_threshold).

Several border treatments can be chosen for filtering via the parameter MarginMarginMarginMarginmarginmargin:

When using the MaskTypeMaskTypeMaskTypeMaskTypemaskTypemask_type 'square'"square""square""square""square""square" with RadiusRadiusRadiusRadiusradiusradius 1 or 2 (resulting in a 3x3 or 5x5 pixel filter mask) and the border treatment 'mirrored', median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image can be executed on OpenCL devices.

For an explanation of the concept of smoothing filters see the introduction of chapter Filters / Smoothing.


median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image uses an algorithm with a runtime per pixel that depends on the mask height 2 * RadiusRadiusRadiusRadiusradiusradius + 1. Therefore, median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image is slower than median_rectmedian_rectMedianRectMedianRectMedianRectmedian_rect for square masks with a large mask height. The precise mask height for which median_rectmedian_rectMedianRectMedianRectMedianRectmedian_rect will become faster than median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image depends on the computer architecture (processor type, availability of SIMD instructions like SSE2 or MMX, cache size and throughput, memory throughput). Typically, this is the case for mask heights > 15, but can also be the case only for larger mask sizes, e.g., if SIMD instructions are unavailable and memory throughput is low.

Furthermore, it should be noted that median_rectmedian_rectMedianRectMedianRectMedianRectmedian_rect uses a recursive implementation, which internally computes the filter response on the smallest enclosing rectangle of the domain of the input image. Therefore, if the domain of the input image only covers a small fraction of the smallest enclosing rectangle, it can happen that median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image is faster than median_rectmedian_rectMedianRectMedianRectMedianRectmedian_rect even for larger mask heights.

Due to performance reasons, the input ImageImageImageImageimageimage is not checked whether it contains NaNs. Using an input image with NaNs crashes HALCON.

Note that filter operators may return unexpected results if an image with a reduced domain is used as input. Please refer to the chapter Filters.

Execution Information


ImageImageImageImageimageimage (input_object)  (multichannel-)image(-array) objectHImageHObjectHImageHobject (byte* / int2* / uint2* / int4* / real*) *allowed for compute devices

Image to be filtered.

ImageMedianImageMedianImageMedianImageMedianimageMedianimage_median (output_object)  (multichannel-)image(-array) objectHImageHObjectHImageHobject * (byte / int2 / uint2 / int4 / real)

Filtered image.

MaskTypeMaskTypeMaskTypeMaskTypemaskTypemask_type (input_control)  string HTuplestrHTupleHtuple (string) (string) (HString) (char*)

Filter mask type.

Default value: 'circle' "circle" "circle" "circle" "circle" "circle"

List of values: 'circle'"circle""circle""circle""circle""circle", 'square'"square""square""square""square""square"

List of values (for compute devices): 'square'"square""square""square""square""square"

RadiusRadiusRadiusRadiusradiusradius (input_control)  integer HTupleintHTupleHtuple (integer) (int / long) (Hlong) (Hlong)

Radius of the filter mask.

Default value: 1

List of values (for compute devices): 1, 2

Suggested values: 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 15, 19, 25, 31, 39, 47, 59

Typical range of values: 1 ≤ Radius Radius Radius Radius radius radius ≤ 4095

Minimum increment: 1

Recommended increment: 2

MarginMarginMarginMarginmarginmargin (input_control)  string HTupleUnion[int, float, str]HTupleHtuple (string / integer / real) (string / int / long / double) (HString / Hlong / double) (char* / Hlong / double)

Border treatment.

Default value: 'mirrored' "mirrored" "mirrored" "mirrored" "mirrored" "mirrored"

List of values (for compute devices): 'mirrored'"mirrored""mirrored""mirrored""mirrored""mirrored"

Suggested values: 'mirrored'"mirrored""mirrored""mirrored""mirrored""mirrored", 'cyclic'"cyclic""cyclic""cyclic""cyclic""cyclic", 'continued'"continued""continued""continued""continued""continued", 0, 30, 60, 90, 120, 150, 180, 210, 240, 255

Example (HDevelop)

read_image (Image, 'fabrik')
median_image (Image, Median, 'circle', 3, 'continued')


For each pixel: O(2 * RadiusRadiusRadiusRadiusradiusradius + 1).


If the parameter values are correct the operator median_imagemedian_imageMedianImageMedianImageMedianImagemedian_image returns the value TRUE. The behavior in case of empty input (no input images available) is set via the operator set_system('no_object_result',<Result>)set_system("no_object_result",<Result>)SetSystem("no_object_result",<Result>)SetSystem("no_object_result",<Result>)SetSystem("no_object_result",<Result>)set_system("no_object_result",<Result>). If necessary, an exception is raised.

Possible Predecessors


Possible Successors

thresholdthresholdThresholdThresholdThresholdthreshold, dyn_thresholddyn_thresholdDynThresholdDynThresholdDynThresholddyn_threshold, regiongrowingregiongrowingRegiongrowingRegiongrowingRegiongrowingregiongrowing


median_rectmedian_rectMedianRectMedianRectMedianRectmedian_rect, rank_imagerank_imageRankImageRankImageRankImagerank_image, rank_rectrank_rectRankRectRankRectRankRectrank_rect

See also

gray_erosion_rectgray_erosion_rectGrayErosionRectGrayErosionRectGrayErosionRectgray_erosion_rect, gray_dilation_rectgray_dilation_rectGrayDilationRectGrayDilationRectGrayDilationRectgray_dilation_rect, gray_erosion_shapegray_erosion_shapeGrayErosionShapeGrayErosionShapeGrayErosionShapegray_erosion_shape, gray_dilation_shapegray_dilation_shapeGrayDilationShapeGrayDilationShapeGrayDilationShapegray_dilation_shape, gray_erosiongray_erosionGrayErosionGrayErosionGrayErosiongray_erosion, gray_dilationgray_dilationGrayDilationGrayDilationGrayDilationgray_dilation


T.S. Huang, G.J. Yang, G.Y. Tang; “A Fast Two-Dimensional Median Filtering Algorithm”; IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 27, no. 1, pp. 13-18, 1979.
R. Haralick, L. Shapiro; “Computer and Robot Vision”; Addison-Wesley, 1992, pp. 318-320.