Computers in Microscopy

A Glossary of Image Processing Terms

Aliasing - the problem of using too coarse a grid to sample a continuous function. Fluctuations on a scale finer than that of the grid are distinguishable from coarser detail. The artificial low spatial frequency components cannot be separated from real structure in the original image. If digitised grid intensities are then displayed as discrete pixels, high frequency (sharp) information may appear to be present in the digitised image, particularly along smooth edges which appear "stairstepped" or "jagged". Special "anti-aliasing" techniques can be used to suppress the jagged appearance of the display but these do not correct for any undersampling during the acquisition of the data.

Analogue - an analogue signal is capable of having a continuously varying value. Contrast with digital.

Analogue to digital converter - a device that senses an analogue signal and converts it into a proportional representation in digital form.

Artefact(ing) - Misinterpreted information from a JPEG or compressed image. Colour faults or line faults that have a visible negative impact on the image.

Bi-level image - another term for a binary image.

Binary (sometimes Bit) image - an image where pixel intensities take only two possible values, either zero or a specific non-zero value (typically unity or full-scale).

Binary images take much less space to store in the image memory of the computer or on a disc since each pixel can be represented by one bit (see bit), and they are convenient for making measurements, e.g. area or perimeter of a feature. Often, a binary image may be shown as white (for 1) and black (for 0).

Binary image operators - calculations on a binary image. These include:-

Erosion - the removal of a layer of pixels around a feature. Small features may be totally removed, separated into their components.

Dilation - The reverse of erosion; adding a layer of pixels around a feature. Features which are fragmented can be rejoined.

Opening - Erosion followed by dilation may separate features which were linked by a narrow neck. Small features can be lost, whilst leaving the size of resultant features approximately unchanged.

Closing - Dilation followed by erosion can be used to fill in small holes or cracks. In order to remove all irregularities, both protuberances and holes, opening followed by closing can be performed.

Skeletonisation - Similar to erosion, except that the process stops when the feature is only one pixel wide. A broad line would be reduced a thin line, although the irregularities in the original line may produce many "whiskers" branching off the resultant thin line.

Bit - the smallest unit of digital information, having the value "0" or "1" only. The term is a contraction from ‘binary’ and ‘digit’. A binary image is a "1-bit" image, whilst a grey image may be stored as a "6-bit" image since a traditional tube TV camera can distinguish about 64 grey values (26 = 64). Note that modern CCD cameras (see below) can distinguish 1000 or more grey values.

Bit Depth - this refers to the grey scale (or colour scale) adopted for an individual pixel. A pixel with 8 bits per colour gives a 24 bit image. (8 bits x 3 colours is 24 bits.)

Bitmap - the method of storing information that maps an image pixel, bit by bit. There are many bitmapped file formats, .bmp, .pcx, .pict, .pict-2, tiff, .tif, .gif (89a), and so on. Most image files are bit mapped. This type of file often gives rise to the effect known as ‘jaggies’ - when examined closely the line of pixels that create edges can be seen. Bitmap images are used by all computers. For example, the desktop or screen information for MS Windows machines uses .bmp files, while the Macintosh uses .pict files.

Blob - a term sometimes used to describe a region of connected pixels of the same type.

Byte - a unit of digital information, consisting of eight bits. A byte can have 28 or 256 different combinations of bits. Much computer hardware is geared to handling information in bytes. Coincidentally, in many applications, a byte is sufficient to represent the information about the grey level of a pixel.

Kilobyte - 210 or 1024 bytes, written KB.

Megabyte - 220 bytes or 1024 kilobytes, written MB. Both terms are used to refer to size of files or media such as hard drives where capacities are too great to be expressed directly in bytes. Refers to amount of information in a file, or how much information can be contained on a Hard Drive or Disk.

CCD camera - a video camera containing a Charge-Coupled Device, which is a semiconductor light sensor in the form of an integrated circuit. CCDs are smaller, more robust, and offer better performance than older video cameras which make use of a vidicon tube as the sensor. In their normal condition, CCDs are greyscale devices. To create colour a colour pattern is laid down on the sensor pixels, using an RGBG colour mask (Red, Green, Blue, and Green). The extra Green is used to create contrast in the image. The CCD Pixels respond independently to coloured light and pass their electronic responses into a shift register for storage. CCDs are analogue sensors; digitising happens when the electronic signal from the shift register is passed to the input of an Analogue to Digital Converter (ADC). The ADC converts the analogue signal to a sequence of digital numbers.

Chrominance - the colour portion of a video signal.

Chain-code - a method of coding information about the position of the boundary of an object, in which the information held is is simply the direction in which one must move in order to reach the next pixel on the boundary. In many cases this is a very efficient way of encoding boundary information.

CMYK - Cyan, Magenta, Yellow, Black. These are the printed colours typically used to create colour prints. Most colour printers - Ink-Jet, Laser, Dye-Sublimation, Thermal, and Crayon printers - use these as their printer colours. This is one of the colour management problems for computers, since converting RGB files to CMYK files may cause unwanted colour shifts. When used by a printer, CMYK is also known as reflective color since it is printed on paper or reflective films.

Compression - techniques used by imaging devices and software to reduce image storage requirements without objectionably affecting the appearance of the image.

Digital - with discrete values (in contrast with analogue). Digital data can be handled without errors, whereas manipulation of analogue data will always entail some finite degree of error, however small.

Digitisation - conversion of analogue information to digital form. For example, the output of an SEM electron detector requires digitisation in both spatial and amplitude terms before it can be processed using digital computing methods.

Digitising tablet - a tablet connected to a computer or image analyser on which the features or interest in a specimen can be drawn using a hand-held cursor or electronic "pen". The x,y coordinates of the cursor are processed by the host computer and may be used to calculate the area, perimeter, centre or gravity, orientation, etc. of the feature.

DQE - detection quantum efficiency, i.e. "efficiency of recording quanta"

Dynamic range - the ratio between the brightest and dimmest grey level acceptable to an imaging system.

EPS - Encapsulated Postscript, a computer file standard set by Adobe for printers, which represents a mathematical definition of shapes, lines, color and space. This standard is one of the most accurate ways to define a font or image, but creates files of very considerable size. EPS files also add page description information to the files. Although forms of EPS are used on all computer, not all postscript files are the same, nor are they transferable between programs. EPSF is an IBM file type generally, EPSP is usually found on Macintosh, and there are many other variations of each type.

False colour (Pseudo colour) - an artificial display where colours are used to represent different ranges of intensity. The choice of colours is often arbitrary so to this extent is "false". However, there may be genuine relationship between intensity and some physical parameter (e.g. atomic number), in which case "colour coding" is a more appropriate term. A grey image may be displayed in false colour to highlight particular features.

Feature - part of an image which can be isolated from the remainder by some means. For example, a "particle" can often be distinguished by having a generally brighter intensity than its surroundings. Separation of an image into features is referred to as "segmentation".

Filter - a device, method or program algorithm that separates signals or other data based upon certain specified criteria

Frame grabber - an input device that picks up an image from a video camera and digitises the analogue signal into a defined number of bits per pixel, and transmits it to some form of digital memory.

GIF - Graphic Interchange Format, originally used by CompuServe for storage of graphical images in lossless compressed form. This was developed for use with images of up to 256 colours or shades (up to 8 bits) and is effective with low-noise or graphical images. This format was widely used in the late 80s for image transfer. GIF 89 is the most recent GIF standard and allows multiple images within one file, selection of area for transparency as well as other features. The primary use of this format has been on the Internet. However, there has since been dispute over the ownership of the embedded compression technology, and other formats are now more widely used.

Grey image - often known as a grey scale image, or grey level image: an image in which the regions or points may take on a range of values or grey levels (distinguished from a binary image in which only two are allowed). To match a conventional 625-line TV camera a typical grey image consists of 512 lines with 512 pixels on each line, i.e. 262144 pixels.

Grey level - the amplitude or energy level associated with a region or point of an image.

Grey level histogram - in digital image processing, "histogram" refers to the distribution showing the number of pixels which have a particular intensity value.

Hue - the name of a colour such as red or blue. The hue of a colour is its dominant wavelength.

HSI - hue, saturation, intensity. An abbreviation for all of a colour's characteristics: hue (the pigment); the saturation (the amount of pigment); and intensity (the amount of white included). All colours can be defined by expressing these characteristics in percentages.

Image - a two-dimensional representation of the response from some sensing device, responding typically but not necessarily to light or other radiation. The process of digitisation produces a digital image consisting of a table of numeric values which describe the relative positions and values of the original responses.

Image analysis - an operation or set of operations designed to yield a numerical or logical result from an image which can be expressed in non-image terms. Image analysis is often preceded by image processing.

Image arithmetic - although a digital image can be thought of as a 2-D array or "matrix" of values, mathematical equations involving images do not always follow the rules of matrix algebra. Thus, if i and j are the row and column of a particular pixel in an image A with intensity A(i,j), then the sum of images A and B, A+B is an image C where: -

C(i,j) = A(i,j) + B(i,j)

for all pixels. However the product of the two images A and B, A*B is an image C where:-

C(i,j) = A(i,j) * B(i,j)

for all pixels, rather than the formal definition of a matrix product.

Although storage of intensities in "integer" format, with discrete values, is normally adequate for representing the original image, intermediate values during calculation may fall outside the range even though the final result is still within range (e.g. during a Fourier transform). Care has to be taken with integer arithmetic to preserve precision through an arithmetic operation. In these cases floating-point arithmetic may be used.

Image processing - operations designed to make an image more useful. Image processing is often regarded as embracing image analysis, but in this course we tend to regard image processing as the set of methods used to enhance the image prior to image analysis.

Intensity - a term which is used rather loosely in this context, mainly to mean the number stored for a given pixel in the digitised image.

JPEG - a standard for still image compression. Devised by the Joint Photographic Experts Group and sanctioned by the International Standards Organisation (ISO) and the CCITT. There are several implementations of the JPEG standard, some proprietary. JPEG, also known as JFIF takes areas of 8 x 8 pixels and compresses the information using the Discrete Cosine Transform. Parts of the resultant data may be sacrificed if loss in fidelity may be tolerated, leading to a high degree of compression. Other forms of compression may then be applied to present the data in a still more compact form. With high compression the 8 x 8 pixel blocks may become apparent in an objectionable manner. This effect is sometimes described colloquially as ‘blockiness’ - the higher the compression the more prominent the ‘blockiness’.

Kalman filter - a recursive procedure which enables estimation of signals to be determined from successive measurements in time. In the context of image acquisition, this normally refers to the technique of using a single frame store to accumulate an image which is close to the true average of all acquired frames since the store was last initialised. For a given pixel, the true average over all frames often represents the best estimate of true intensity for that pixel.

Lab Colour - L*a*b* is a color model developed by the Centre Internationale dŽEclairage (CIE). These standards are internationally accepted standards for colorimetric measurements. The Lab model, like other CIE colour models, defines colour values mathematically, in a device independent manner. Lab colour is consistent, regardless of the device producing the colour.

Look-up table - a means whereby the digitised video signal may be altered to produce a more readily interpreted display. Instead of displaying an image so that the brightness of each pixel is proportional to its numerical value, the values in the stored digital image can be converted on display to a different value using the look-up table. A grey image can therefore be displayed as a negative or part of the range of grey values expanded without changing the stored data. By giving different tables for different colours the original grey image can be displayed in false colour. A look-up table can also be applied to the input from the TV camera.

Lossless compression - image and data-compression applications and algorithms, such as Huffman Encoding, that reduce the number of bits a picture would normally take up without sacrificaing any fidelity.

Lossy compression - methods of image compression, such as JPEG, that reduce the size of an image by sacrificing some pictorial information.

Luminance - the black and white or brightness portion of a video signal.

Mapping - the mathematical conversion of one set of numbers into a different set based upon some transformation.

Monochrome image - an image displayed in shades of a single colour, normally grey.

Neighbourhood (or local) operations - operations on an image using the neighbourhood of each pixel to calculate a new value for that pixel. Typically a 3 x 3 or 5 x 5 group of pixels is used to "smooth" an image or enhance some aspect of the image, such as the edge of a feature.

Object - in a binary image, a region in which all the pixels are connected (according the the connectivity rules that apply in the image analysis system) , and which is surrounded by pixels of the other type.

Object-measurements - measurements which relate to the individual objects in the image, and which can be combined with others on an object-by-object basis to give other information - for example, roundness or aspect ratio.

Overlay - superimposition of a binary image, graphics or text over a grey image, using colour or good contrast to differentiate between the two images.

Pointset - a set of connected pixels which can be represented in an efficient way, allowing very fast access to the data they represent.

Parallax - the difference in position of the same feature between two images recorded under different conditions (e.g. sample tilt).

Pixel (or picture element) - normally refers to one of the grid points on the 2-D grid or the corresponding element in the 2-D array of intensity values representing the digitised image. 640 x 480 is the pixel resolution of most VGA monitors. Pixels are typically square in computers, but may be rectangular in video applications.

Pixellation - The stair-stepped appearance of a curved or angled line in digital imaging. The smaller the pixels, and the greater their number, the less apparent the "pixellation" of the image. Also known colloquially as the "jaggies".

Plug-In - a program architecture was first popularized by Adobe Photoshop and is now a de facto standard for all major imaging programs. Compared with Twain it allows more flexibility in design so that acquisition, export and other specific tasks can be performed within a software application. This is the preferred choice of operation in the Macintosh computers. Plug-In ideology has spread to other applications like Netscape Navigator, Macromedia Director, and so on. Not all plug-ins work with all products and specific interfaces are required for different types of software. Adobe’s software has become a de facto standard for image editing software and graphic illustration software.

Pointset - a set of connected pixels which can be used to describe that feature, e.g. the boundary of a feature.

Precision - the extent to which a result can be reproduced in successive experiments as opposed to the difference between the result and the "true" value.

RAM/ROM - Random access memory, and Read-only memory. Memory is a vital element of any computing or image processing system, in which program information, image and other data can be stored. Any of the contents of RAM or ROM can be freely accessed at high speed, in contrast to serial storage devices like magnetic tape (and, to some extent, magnetic disk) in which the data stored can only be accessed in a fixed order. RAM can be read from and written to, while ROM (Read-only memory) can only be read from. ROMs are usually programmed by the manufacturer of the system, and their content is permanent and cannot normally be changed by the user.

SRAM - Static RAM, is the fastest but most expensive type of RAM, often found in framestores, some Printers, and in PCMCIA Type I Cards. Compare with DRAM (dynamic RAM), which is slower and cheaper, and is most often seen as the expandable RAM used in the computer for memory.

Remote sensing - the acquisition and interpretation of satellite and aerial images, seismographic and other related responses, often for geographical or land-use studies.

Resolution - the ability to discriminate closely spaced objects in an image. The resolution of an image analysis system may be governed by electronic, optical or other considerations. The density of the sampling points affects the amount of detail visible in an image.

RGB - an additive colour model that forms colours by mixing various ratios of red green and blue. Computer monitors and digital cameras use these primary colours to create all the hues seen on the monitor and saved in files.

Saturation - designates the purity of a colour or how much the colour is diluted by white. Red is a highly saturated colour. Pink has the same hue but a lower saturation.

S/N ratio (or signal to noise ratio) - the ratio between the level of signal (or significant information) and the level of noise. Often used rather loosely to indicate the extent of information of interest as opposed to spurious detail in an image. In this context, noise is predominantly the fluctuations due to counting statistics.

Screen widths - an imaginary screen can be defined for display of a scanned image where at a given magnification:-

Width of field on specimen = Screen width x Magnification.

Segmentation - the process of separating an image into the various features, discriminating those parts of an image which are of interest from those which are not (the 'background'). The simplest type of segmentation is thresholding.

Shape factor - a number derived from two or more measurements which indicates something particular about the shape of the feature. There are many different shape factors, e.g. Maximum projection / Minimum projection will be large for needle-like features but near unity for circular objects. Measurements are combined to form a factor which is dimensionless and thus independent of the size of feature, for example Area / (Maximum projection)2.

Smoothing - a smoothed image is one where the intensity at a pixel has been replaced by a value based on the pixel intensities in the immediate neighbourhood. This can be a simple average over a 3 x 3 area or a weighted average where smoothing coefficients are used (often arbitrarily!) to weight according to the distance from the central pixel (e.g. "Gaussian" smoothing). The median of the neighbourhood can also be used. Smoothing changes spiky statistical fluctuations into blobby artefacts and can only be used reliably where an independent measure of noise is available.

Spatial frequency - the rate of change of pixel intensity in an image. Usually regarded as a two-dimensional quantity, with both horizontal and vertical components.

Stereology - the study and measurement of the average properties of a three-dimensional structure from measurements made on two-dimensional sections. Depending on the shape and distribution of features on a section it may be possible to calculate the properties of the original specimen, e.g. Volume fraction of a component = Area fraction of that component measured on sections. Surface area estimation from linear measurements is not so easy.

Thresholding - the simplest process for converting a grey scale image to a binary image by comparing each pixel value to a threshold value. In one implementation those pixels exceeding the threshold become white, while the remainder become black.

Threshold - a grey level value which can be used to distinguish objects or regions of interest from the background. In many classes of image, a single threshold value may not be sufficient for satisfactory segmentation.

TIFF - Tagged Image File Format. A bit map file format for describing and storing colour and grey scale images.

TV scan rate - in conventional (CCIR - European standard) TV images, the scanning beam traverses the full width of the field of view every 64 m sec, and covers the even-numbered lines within the field in 20 msec, and the odd lines in a further 20 msec. A complete image frame consists of interlaced even and odd lines (referred to as the even and odd fields). Hence, it takes 40 msec to build an entire image frame.

Twain - An interface for image acquisition developed by a consortium of software developers as a standard for communications between scanners, imaging devices (and more recently digital cameras) and the computer. Twain provides for easy import or acquisition of images from a range of devices into software applications that support the standard. This is a widely used interface on the MS Windows platform.

WYSIWYG - an acronym which stands for "what you see is what you get". This refers to the possibility with modern software of viewing on the screen accurate representations of images which can be printed out. The term came into vogue with the development of word processing software, but is often used in the context of packages for DTP or image manipulation. Pronounced "WizzyWig".

D M Holburn

Created September 1996
Updated September 1997
Copnverted to HTML September 1998