Properties generated by OpenSlide¶

openslide.background-color

openslide.barcode

openslide.bounds-height

openslide.bounds-width

openslide.bounds-x

openslide.bounds-y

openslide.comment

openslide.icc-size

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

openslide.quickhash-1

openslide.vendor

Properties for TIFF-based formats¶

tiff.Artist

tiff.Copyright

tiff.DateTime

tiff.DocumentName

tiff.HostComputer

tiff.ImageDescription

tiff.Make

tiff.Model

tiff.ResolutionUnit

tiff.Software

tiff.XPosition

tiff.XResolution

tiff.YPosition

tiff.YResolution

Vendor-specific properties¶

A list of vendor-specific properties can be found on the pages for each vendor format[1].

Vendor Documentation¶

Documentation PDF[2]

Detection¶

Aperio slides are stored in single-file TIFF format. OpenSlide will detect a file as Aperio if:

Relevant TIFF tags¶

Extra data stored in ImageDescription¶

For tiled images, the ImageDescription tag contains some dimensional downsample information as well as what look like offsets. Additionally, vertical line-delimited key-value pairs are stored, in at least the full-resolution image. A key-value pair is equals-delimited. These key-values are stored as properties starting with “aperio.”. Currently, OpenSlide does not use any of the information present in these key-value fields.

TIFF Image Directory Organization¶

Digital Slides and Third-Party Data Interchange[2] page 14:

The first image in an SVS file is always the baseline image (full resolution). This image is always tiled, usually with a tile size of 240x240 pixels. The second image is always a thumbnail, typically with dimensions of about 1024x768 pixels. Unlike the other slide images, the thumbnail image is always stripped. Following the thumbnail there may be one or more intermediate “pyramid” images. These are always compressed with the same type of compression as the baseline image, and have a tiled organization with the same tile size.

Optionally at the end of an SVS file there may be a slide label image, which is a low resolution picture taken of the slide’s label, and/or a macro camera image, which is a low resolution picture taken of the entire slide. The label and macro images are always stripped.

JPEG 2000 (compression types 33003 or 33005)¶

Some Aperio files use compression type 33003 or 33005. Images using this compression need to be decoded as a JPEG 2000 codestream. For 33003: YCbCr format with no MCT, possibly with a chroma subsampling of 4:2:2. For 33005: MCT, which OpenJPEG decodes into RGB. Note that the TIFF file may not encode the colorspace or subsampling parameters in the PhotometricInterpretation field, nor the YCbCrSubsampling field, even though the TIFF standard seems to require this. The correct subsampling can be found in the JPEG 2000 codestream.

ICC Profiles¶

The slide ICC profile is taken from the ICC Profile tag of the first image.

Associated Images¶

There are up to three stripped images: a thumbnail image, which is always the second image in the file; and label and macro images at the end of the file, with subfile types 1 and 9 respectively.

label

macro

thumbnail

Known Properties¶

All key-value data encoded in the ImageDescription TIFF field is represented as properties prefixed with “aperio.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Aperio/[3]

Vendor Documentation¶

Documentation PDF[4]

Detection¶

ARGOS slides are stored in single-file BigTIFF format. OpenSlide will detect a file as ARGOS if:

Relevant TIFF tags¶

Metadata XML¶

The first TIFF directory includes metadata XML in TIFF tag 65000. The field type is ASCII, but in our samples the value is not NUL-terminated as required by TIFF, causing libtiff to warn about it.

TIFF Image Directory Organization¶

The TIFF file contains the image pyramid of the lowest focal plane, followed by each higher plane in succession. Images are sparse, with missing tiles represented by a zero TileOffset and a zero TileByteCount.

The last two TIFF directories are the thumbnail and macro images, which are both stripped.

Associated Images¶

thumbnail

macro

Known Properties¶

All simple key-value data encoded in the metadata XML is represented as properties prefixed with “argos.”.

openslide.barcode

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Argos/[5]

Format Documentation¶

DICOM Whole Slide Imaging[6]

Detection¶

OpenSlide will detect a file as DICOM if:

File Organization¶

OpenSlide uses libdicom[7] to parse and load DICOM files.

OpenSlide non-recursively scans the containing directory for other DICOM files with the same Series Instance UID attribute value as the specified file. It uses the Image Flavor (value 3 of the Image Type attribute) of each matching file to determine the file’s role within the whole slide image:

A slide level can be spread across multiple files that share a Concatenation UID.

If multiple files in the directory have the same SOP Instance UID, the extra copies are ignored.

A file’s Dimension Organization Type can be TILED_FULL, TILED_SPARSE, or 3D. These transfer syntaxes and photometric interpretations are supported:

ICC Profiles¶

The slide ICC profile is taken from the ICC Profile attribute of the highest-resolution image. Associated image ICC profiles are taken from the ICC Profile of the associated image.

Associated Images¶

Known Properties¶

Many DICOM attributes are represented hierarchically as properties prefixed with “dicom.”.

openslide.barcode

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/DICOM/[8]

Detection¶

OpenSlide will detect a file as Hamamatsu if:

or if:

Overview¶

The Hamamatsu format has three variants. VMS and VMU consist of an index file, 2 or more image files, and (in the case of VMS) an “optimisation” file. NDPI consists of a single TIFF-like file with some custom TIFF tags. VMS and NDPI contain JPEG images; VMU contains NGR images (a custom uncompressed format).

Multiple focal planes are ignored, only focal plane 0 is read.

JPEG does not allow for files larger than 65535 pixels on a side. In VMS, multiple JPEG files are used to encode large images. To avoid having many files, VMS uses close to maximum size (65K by 65K) JPEG files. NDPI, instead, stuffs large levels into a single JPEG and sets the overflowed width/height fields to 0.

Unfortunately, JPEG provides very poor support for random-access decoding of parts of a file. To get around this, JPEG restart markers are placed at regular intervals, and these offsets are specified in the optimisation file (in VMS) or in a TIFF tag (in NDPI). With restart markers identified, OpenSlide can treat JPEG as a tiled format, where the height is the height of an MCU row, and the width is the number of MCUs per row divided by the restart marker interval times the width of an MCU. (This often leads to oddly-shaped and inefficient tiles of 4096x8, for example.)

Unfortunately, the VMS optimisation file does not give the location of every restart marker, only the ones found at the beginning of an MCU row. It also seems that the file ends early, and does not give the location of the restart marker at the last MCU row of the last image file.

Thus, the optimisation file can only be taken as a hint, and cannot be trusted. The entire set of JPEG files must be scanned for restart markers in order to facilitate random access. OpenSlide does this lazily as needed, and also in a background thread that runs only when OpenSlide is otherwise idle.

The VMS map file is a lower-resolution version of the other images, and can be used to make a 2-level JPEG pyramid. JPEG also allows for lower-resolution decoding, so further pyramid levels are synthesized from each JPEG file.

VMS File¶

The .vms file is the main index file for the VMS format. It is a Windows INI-style key-value pair file, with sections. Only keys in the Virtual Microscope Specimen group are read by OpenSlide.

Here are known keys from the file:

VMU File¶

The .vmu file is the main index file for the VMU format. Only keys in the Uncompressed Virtual Microscope Specimen group are read by OpenSlide.

Here are known keys from the file:

NDPI File¶

NDPI uses a TIFF-like structure, but libtiff cannot read the headers of an NDPI file. This is because NDPI specifies the RowsPerStrip as the height of the file, and after doing out the multiplication, this typically overflows libtiff and it refuses to open the file. Also, the TIFF tags are not stored in sorted order, and ASCII values are always stored out-of-line even when TIFF would store them inline.

NDPI stores an image pyramid in TIFF directory entries. In some files, the lower-resolution pyramid levels contain no restart markers. The macro image, and sometimes an active-region map, seems to come last.

JPEG files in NDPI are not necessarily valid. If ImageWidth or ImageHeight exceeds the JPEG limit of 65535, then the width or height as stored in the JPEG file is 0. libjpeg will refuse to read the header of such a file, so the JPEG data stream must be altered when fed into libjpeg.

Conversely, on some pyramid levels without restart markers, the TIFF ImageWidth and ImageHeight may not match the JPEG dimensions due to inconsistent rounding.

NDPI is based on the classic TIFF format, which does not support files larger than 4 GB. However, NDPI files can be larger than 4 GB. NDPI handles this by storing the high-order bits of file offsets beyond the ends of standard TIFF structures. TIFF puts next-directory offsets (in the TIFF header and at the end of each directory) at the ends of their parent data structures, so NDPI simply extends these values to 64-bit quantities. The high-order bits of directory entry Value Offsets are stored in an array immediately after the end of each TIFF directory, 4 bytes per directory entry. Values that would normally be stored inline in the Value Offset (not ASCII, as mentioned above) are still stored inline, even if the high 32 bits are non-zero (e.g. StripOffsets and StripByteCounts, which always have a Count of 1 in NDPI), and other values are still stored out-of-line.

Here are the observed TIFF tags:

Optimisation File (only for VMS)¶

The optimisation file contains a list of 32- (or 64- or 320- ?) bit little endian values, giving the file offset into an MCU row, each offset starts at a 40-byte alignment, and the last row (of the entire file, not each image) seems to be missing. The offsets are all packed into 1 file, even with multiple images. The order of images is left-to-right, top-to-bottom.

Map File (only for VMS/VMU)¶

The VMS map file is a standard JPEG file. Its restart markers (if any) are not included in the optimisation file. The VMU map file is in NGR format. This file can be used to provide a lower-resolution view of the slide.

Image Files (only for VMS/VMU)¶

These files are given by the VMS/VMU ImageFile keys. They are assumed to have a height which is a multiple of the MCU height. They are assumed to have a width which is a multiple of MCUs per row divided by the restart interval.

For VMS, these files are in JPEG, for VMU they are in NGR format.

NGR Format¶

The NGR file contains uncompressed 16-bit RGB data, with a small header. The files we have encountered start with GN, two more bytes, and then width, height, and column width in little endian 32-bit format. The column width must divide evenly into the width. Column width is important, since NGR files are generated in columns, where the first column comes first in the file, followed by subsequent files. Columns are painted left-to-right.

At offset 24 is another 32-bit integer which gives the offset in the file to the start of the image data. The image data we have encountered is in 16-bit little endian format.

Associated Images¶

macro

Known Properties¶

All key-value data stored in the VMS/VMU file, and known tags from the NDPI file, are encoded as properties prefixed with “hamamatsu.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

NDPI format

VMS format

ImHex Patterns¶

<listitem>VMS optimisation file[11] </listitem><listitem>VMU NGR file[12] </listitem><listitem>NDPI file[13] </listitem>.SH "HURON FORMAT"

Format

File extensions

OpenSlide vendor backend

Detection¶

Huron slides are stored in single-file TIFF format. OpenSlide will detect a file as Huron if:

Relevant TIFF tags¶

Extra data stored in ImageDescription¶

The ImageDescription tag of the first TIFF directory contains newline-delimited key-value pairs. A key-value pair is equals-delimited with extra padding spaces. These key-values are stored as properties starting with “huron.”. Currently, OpenSlide does not use any of the information present in these key-value fields.

TIFF Image Directory Organization¶

The directory organization follows the same pattern as the Aperio format[14].

Associated Images¶

There are up to three stripped images: a thumbnail image, which is always the second image in the file; and label and macro images at the end of the file, with subfile types 1 and 9 respectively.

label

macro

thumbnail

Known Properties¶

All key-value data encoded in the ImageDescription TIFF field is represented as properties prefixed with “huron.”.

openslide.mpp-x

openslide.mpp-y

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Huron/[15]

Detection¶

Leica slides are stored in single-file BigTIFF format. OpenSlide will detect a file as Leica if:

Relevant TIFF tags¶

File Organization¶

The ImageDescription tag of the first TIFF directory contains an XML document that defines the structure of the slide.

Leica slides are structured as a collection of images, each of which has multiple dimensions (pyramid levels). The collection has a size, and images have a size and position, measured in nanometers. Each dimension has a size in pixels, an optional focal plane number, and a TIFF directory containing the image data. Fluorescence images have different dimensions (and thus different TIFF directories) for each channel. OpenSlide currently rejects fluorescence images and ignores focal planes other than plane 0.

Brightfield slides have at least two images: a low-resolution macro image and one or more main images corresponding to regions of the macro image. The macro image has a position of (0, 0) and a size matching the size of the collection. Fluorescence slides can have two macro images: one brightfield and one fluorescence.

The slide provides enough information to composite the various images, including the macro image, into a single pyramid. However, there are some complications: <listitem>The resolution of the macro image is generally not related to the resolution of the main images by a power of two. </listitem><listitem>Downsampled dimensions are generally downsampled from the next larger dimension by a factor of 4, but main images can be scanned with distinct objectives that may differ by only a factor of 2. </listitem>.PP Thus, in general, the images in a collection cannot be rendered into a unified pyramid without scaling the original pixel data. OpenSlide does not attempt to do this. Instead, OpenSlide omits the macro image from the pyramid and refuses to open slides whose main images have inconsistent resolutions.

Associated Images¶

macro

Known Properties¶

leica.aperture

leica.barcode

leica.creation-date

leica.device-model

leica.device-version

leica.illumination-source

leica.objective

openslide.barcode

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Leica/[17]

Detection¶

OpenSlide will detect a file as MIRAX if:

Overview¶

MIRAX can store slides in JPEG, PNG, or BMP formats. Because JPEG does not allow for large images, and JPEG and PNG provide very poor support for random-access decoding of part of an image, multiple images are needed to encode a slide. To avoid having many individual files, MIRAX packs these images into a small number of data files. The index file provides offsets into the data files for each required piece of data.

The camera on MIRAX scanners takes overlapping photos and records the position of each one. Each photo is then split into multiple images which do not overlap. Overlaps only occur between images that come from different photos.

To generate level n + 1, each image from level n is downsampled by 2 and then concatenated into a new image, 4 old images per new image (2 x 2). This process is repeated for each level, irrespective of image overlaps. Therefore, at sufficiently high levels, a single image can contain one or more embedded overlaps of non-integral width.

Index File¶

The index file starts with a five-character ASCII version string, followed by the SLIDE_ID from the slidedat file. The rest of the file consists of 32-bit little-endian integers (unaligned), which can be data values or pointers to byte offsets within the index file.

The first two integers point to offset tables for the hierarchical and nonhierarchical roots, respectively. These tables contain one record for each VAL in the HIERARCHICAL slidedat section. For example, the record for NONHIER_1_VAL_2 would be stored at nonhier_root + 4 * (NONHIER_0_COUNT + 2).

Each record is a pointer to a linked list of data pages. The first two values in a data page are the number of data items in the page and a pointer to the next page. The first page always has 0 data items, and the last page has a 0 next pointer.

There is one hierarchical record for each zoom level. The record contains data items consisting of an image index, offset and length within a file, and a file number. The file number can be converted to a data file name via the DATAFILE slidedat section. The image index is equal to image_y * GENERAL.IMAGENUMBER_X + image_x. Image coordinates which are not multiples of the zoom level’s downsample factor are omitted.

Nonhierarchical records refer to associated images and additional metadata. Nonhierarchical data items consist of two values which are usually zero, followed by an offset, length, and file number as in hierarchical records.

Data Files¶

A data file begins with a header containing a five-character ASCII version string, the SLIDE_ID from the slidedat file, the file number encoded into three ASCII characters, and 256 bytes of padding. (In newer slides, the SLIDE_ID and file number are encoded as UTF-16LE, so the second half of each value is truncated away.) The remainder of the file contains packed data referenced by the index file.

Slide Position File¶

The slide position file is referenced by the VIMSLIDE_POSITION_BUFFER.default nonhierarchical section. It contains one entry for each camera position (not each image position) in row-major order. Each entry is nine bytes: a flag byte, the X pixel coordinate of the photo (4 bytes, little-endian, may be negative), and the Y coordinate (4 bytes, little-endian, may be negative). In slides with CURRENT_SLIDE_VERSION ≥ 1.9, the flag byte is 1 if the slide file contains images for this camera position, 0 otherwise. In older slides, the flag byte is always 0.

In slides with CURRENT_SLIDE_VERSION ≥ 2.2, the slide position file is compressed with DEFLATE and referenced by the StitchingIntensityLayer.StitchingIntensityLevel nonhierarchical section. In some such slides, the nonhierarchical record has a second data item which points to a DEFLATE-compressed blob with four bytes of additional metadata per camera position.

Associated Images¶

label

macro

thumbnail

Known Properties¶

All key-value data stored in the Slidedat.ini file are encoded as properties prefixed with “mirax.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

See Also¶

Introduction to MIRAX/MRXS[18]. Note that our terminology has changed since that document was written; where it says “tile”, substitute “image”, and where it says “subtile”, substitute “tile”.

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Mirax/[19]

ImHex Patterns¶

<listitem>Index file[20] </listitem><listitem>Slide position file[21] </listitem>.SH "PHILIPS FORMAT"

Format

File extensions

OpenSlide vendor backend

Detection¶

Philips TIFF files are stored in single-file TIFF or BigTIFF format. OpenSlide will detect a file as Philips if:

File Organization¶

Philips TIFF is an export format. The native Philips format, iSyntax, is a custom multi-file format not currently supported by OpenSlide.

The ImageDescription tag of the first TIFF directory contains an XML document with a hierarchical structure containing key-value pairs. The keys are based on DICOM tags.

The level dimensions given in the TIFF ImageWidth and ImageLength fields, and also in the ImageDescription XML, are merely the TIFF tile size multiplied by the number of tiles in each dimension. Thus, they include the size of the padding in the right-most column and bottom-most row of tiles. Each level typically uses the same tile size but requires a different amount of padding, so the aspect ratios of the levels are inconsistent and the level dimensions are not proportional to the level downsamples. Correct downsamples can be calculated from the levels’ pixel spacings in the XML metadata.

Slides with multiple regions of interest are structured as a single image pyramid enclosing all regions. Slides may omit pixel data for TIFF tiles not in an ROI; this is represented as a TileOffset of 0 and a TileByteCount of 0. When such tiles are downsampled into a tile that does contain pixel data, their contents are rendered as white pixels.

Label and macro images are stored as Base64-encoded JPEGs in the ImageDescription XML. Some slides also store these images as stripped TIFF directories whose ImageDescriptions start with Label and Macro, respectively.

Relevant TIFF tags¶

Associated Images¶

label

macro

Known Properties¶

All key-value data encoded in the DPUfsImport object, in the first DPScannedImage object with a PIM_DP_IMAGE_TYPE of WSI, and in that object’s PixelDataRepresentation objects is represented as properties prefixed with “philips.”.

openslide.barcode

openslide.mpp-x

openslide.mpp-y

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Philips-TIFF/[22]

Detection¶

OpenSlide will detect a file as Sakura if:

File Organization¶

Sakura slides are SQLite 3 database files written by the eXpress Persistent Objects ORM. Tables contain slide metadata, associated images, and JPEG tiles. Tiles are addressed as (focal plane, downsample, level-0 X coordinate, level-0 Y coordinate, color channel), with separate grayscale JPEGs for each color channel. Despite the generality of the address format, tiles appear to be organized in a regular grid, with power-of-two level downsamples and without overlapping tiles. The structure of the file allows scans to be sparse, but it is not clear if this is actually done.

SQL Tables¶

Some irrelevant tables and columns have been omitted from the summary below. DataManagerSQLiteConfigXPO.PP Useful only to get a reference to the unique table. OpenSlide requires this table to contain exactly one row.

SVSlideDataXPO.PP High-level metadata about a slide. OpenSlide assumes this table will contain exactly one row.

SVHRScanDataXPO.PP A single high-resolution scan of a slide from SVSlideDataXPO. OpenSlide assumes this table will contain exactly one row.

SVScannedImageDataXPO.PP Contains associated images other than the thumbnail.

tile.PP This table is most naturally used to map tile coordinates to tile IDs, but is not suitable for individual lookups because it has no useful indexes. In addition, some Sakura slides don’t have it. OpenSlide ignores the table and constructs tile IDs directly from tile coordinates.

Unique table.PP This is the table named by DataManagerSQLiteConfigXPO.TableName. It contains named blobs including the JPEG tile data.

This table stores a variety of blob types.

Header blob.PP The Header blob is a small binary structure containing little-endian integers as follows:

Associated Images¶

label

macro

thumbnail

Known Properties¶

sakura.Creator

sakura.Date

sakura.Description

sakura.DiagnosisCode

sakura.FocussingMethod

sakura.Keywords

sakura.NominalLensMagnification

sakura.ResolutionMmPerPix

sakura.ScanId

sakura.SlideId

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

No public data available. Contact the mailing list[23] if you have some.

Detection¶

Trestle slides are stored in single-file TIFF format. OpenSlide will detect a file as Trestle if:

Relevant TIFF tags¶

Extra data stored in ImageDescription¶

The ImageDescription tag contains semicolon-delimited key-value pairs. A key-value pair is equals-delimited. We use the OverlapsXY and Background Color keys from the ImageDescription, and ignore the rest. All of these values are stored as properties starting with “trestle.”.

TIFF Image Directory Organization¶

The first image in the TIFF file is the full-resolution image. The subsequent images are assumed to be decreasingly sized reduced-resolution images.

Overlaps¶

The OverlapsXY pseudo-field encodes a list of tile overlap values as ASCII.

Example: “64 64 32 32 16 16” (note the initial space).

These values represent the standard overlaps between adjacent tiles in X and Y, in pixels. This example encodes 3 levels worth of overlaps. Further overlaps are assumed to have the value 0.

Individual tile overlaps may differ from the standard overlaps. These individual overlaps are recorded in .tif-Nb files adjacent to the .tif file, where N is the level number. OpenSlide does not read these files, though they have been partially decoded; see issue 21[24] for details.

Associated Images¶

macro

Known Properties¶

All data encoded in the ImageDescription TIFF field is represented as properties prefixed with “trestle.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Trestle/[25]

Detection¶

Ventana slides are stored in single-file BigTIFF format. OpenSlide will detect a file as Ventana if:

ICC Profiles¶

The slide ICC profile is taken from the ICC Profile tag of the highest-resolution level.

Associated Images¶

macro

thumbnail

Known Properties¶

All XML attributes in the iScan element are represented as properties prefixed with “ventana.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Ventana/[26]

Detection¶

OpenSlide will detect a file as Zeiss if:

CZI format¶

CZI slides consist of overlapping image tiles at arbitrary pixel positions. Tiles in level 0 are typically large, e.g. 2056 x 2464 pixels. A slide may or may not include downsampled levels. Slides may include multiple scenes (scan regions), whose pyramids can have different numbers of downsampled levels. Scenes are reflected in the metadata but aren’t very important for rendering the slide, since tile positions are relative to the slide and not the scene.

The storage unit for an image tile is the subblock. A subblock contains pixel data, a small piece of XML metadata (which is ignored by OpenSlide), and a directory entry containing the subblock’s dimensions, pixel format, and compression type. Directory entries are stored alongside individual subblocks, and a second copy of each entry is stored in an array whose location is recorded in the file header.

A dimension represents coordinates along an axis; it contains a short string identifying the dimension, an offset, and for some dimensions, a size in level 0 and in the subblock’s level. Possible dimensions are “X” and “Y” pixel coordinates, a scene identifier “S”, a channel identifier “C” (with value 0 for brightfield), and the tile’s Z-index “M”. Higher Z-indexes should be rendered after lower Z-indexes. The downsample of a subblock can be computed by dividing its X or Y size in level 0 by its size in the subblock’s level.

Pixel data may be compressed in any of several formats. JPEG XR is common. Uncompressed 24 bpp or 48 bpp pixel data is also possible, as well as pixel data compressed with lossless Zstandard[27] compression. CZI’s Zstandard support comes in two flavors, zstd0 and zstd1, with the latter prefixed by an uncompressed header typically three bytes long. A flag in that header indicates the presence of HiLo packing, in which the low bytes of 16-bit color samples are packed together in the first half of the uncompressed buffer and the high bytes in the second half. CZI also permits JPEG and LZW compression, though these do not appear to be common and OpenSlide does not support them.

CZI files can embed arbitrary named attachments, which might be JPEG images, nested CZI files, or other metadata. The Label and SlidePreview attachments are stored as nested CZIs and the Thumbnail attachment is stored as a JPEG.

CZI also includes a large XML document containing scan parameters for the slide. Gamma values.PP Color images generated by at least Axioscan 7 scanners seem not to be gamma corrected before storage, and appear dark. The zeiss.Information.Instrument.Detectors.$camera_id.GammaDefault property (where $camera_id can be read from zeiss.Information.Image.Dimensions.Channels.Channel:0.DetectorSettings.Detector.Id) might be the recommended gamma correction for displaying CZI images. Observed values include 0.45 for Axiocam705c and AxioCamEL color cameras and 1 for Axiocam712m monochrome cameras.

Associated Images¶

label

macro

thumbnail

Known Properties¶

Certain text nodes and attributes from the XML metadata are represented as properties prefixed with “zeiss.”. The available properties are those from the AttachmentInfos, DisplaySetting, Information, and Scaling elements of ImageDocument.Metadata.

openslide.barcode

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Zeiss/[28]

ImHex Patterns¶

<listitem>CZI file[29] </listitem>.SH "GENERIC TILED TIFF FORMAT"

Format

File extensions

OpenSlide vendor backend

Detection¶

OpenSlide will detect a file as generic TIFF if:

TIFF Image Directory Organization¶

The first image in the TIFF file is the full-resolution image. Any other tiled images in the file with the “reduced resolution” bit set are assumed to be reduced-resolution versions of the original.

Starting after OpenSlide 4.0.0, OpenSlide supports generic TIFF files with missing tiles, i.e. tiles with zero bytes of image data. These tiles are rendered as transparent pixels.

ICC Profiles¶

The slide ICC profile is taken from the ICC Profile tag of the highest-resolution level.

Associated Images¶

None.

Known Properties¶

Many TIFF tags are encoded as properties starting with “tiff.”.

Test Data¶

https://openslide.cs.cmu.edu/download/openslide-testdata/Generic-TIFF/[30]