Image. Image processing. Resolution. Intensity histogram. pixel size random uniform pixel distance random uniform

Transcription

1 Image processing Image analogue digital pixel size random uniform pixel distance random uniform grayscale (8 bit): 0 : black 255 : white Color image: R (red), G (green) and B (blue) channels additive combination 2 24, about 16 million colors. object Where? image perception, processing eg. radiology, patology, endoscopy.. and in the normal life. Intensity histogram The different colors (intensities) may be plotted as functions: Resolution The reciprok of the resolution limit PPI (pixels per inch display) DPI (dots per inch printer) line pair / cm Histogram: The frequency of the colors ordered to the pixels in the whole image or in its selected part: frequency The dark tone is dominant in the image onthebaseof the intensity distribution. The whole intensity range is not used! Decreasing resolution: loss of information, subsequent increase of resolution: no more information! intensity

2 Zoom Increasing pixel size. Bitmap : Zoom is convenient to the processing, but there is no more information. Vector graphics: unlimited zoom, but must be described as curves. Addition / Subtraction DSA = Digital Subtraction Angiography Most of the medical images belong to this group. eg. font types or simple geometric figures (triangle, quadrilateral, circle). = (The subtraction of the grayscale values in the case of each pixels in the whole image) negative Transfer functions brightness displayed grayscale(y) y = f(x) original grayscale (x) Distortion It appiers if a 3D structure (eg. organ) is diplayed in 2D. displayed (y) y = x +255 original (x) contrast In an infinite slope, threshold: pixels below the threshold are black, above the threshold are white. gamma undistorted pincushion distortion barrel distortion if there is no any change: input = output

3 The semiconductor detectors are heatsensitive: cooling may Noise: decrease the noise. most common source: detector pepper and salt noise (random black and white pixels) Filtering: Me Effective in the case of pepper and salt noise. X Every periodic signal may be generated as the sum of sinusoidal functions. time function y( t) = ak sin( k ω0 t + Φ k ) k FT Fourier principle inverz FT frequency function Square pulse: base frequency + adequate harmonics ( right frequency and amplitude). base frequency harmonics 1D: f(x) F(u) spatial frequency 2D: f(x,y) F(u,v) 2D spatial frequency image is a 2D the sum median mean Fourier transformation (FT) (spatial domain frequency domain) Fourier transformation (FT) (spatial domain frequency domain) regular irregular frequency domain (repeating fences) spatial domain (house and fences)

4 Fourier reconstruction (frequency domain spatialdomain) The sharp changes in the image may be described exactly by high frequency components. corrected Application of the Fouriertransformation : a) noise filtering striated pattern as a periodic square pulse image Fouriertransformation : b) to find edges symmetrically increasing frequencies to the right and left side Fourier transformation Aliasing (sampling problem) The sampling frequency must be the double of the signal frequency at least! Moiré The ratio of the pixelsize of the detector and the pattern Data loss or false data may appier, if the image should be displayed on that grid (pixels of a detector).

5 overexposed underexposed left side mastectomy Windowing as a transfer function μx μ HU = μ water water 1000 µ: linear attenuation coefficient bone: HU HU HU eg. lung water nipples

6 File types uncompressed, eg. BMP (bitmap) small storage requirements, but a compressed part of the information content Lossless, eg. TIFF, GIF looses Lossy, eg. JPG Digital Imaging Communications in DICOM Medicine image (or series of images) (uniform file type + network communication protocol) patient data Data on imaging mode / settings of the equipment, windowing, radiation doseinthecaseofct, etc.