Computer Vision: 3. Images
3.1 Image
Resizing image
- if we throw away even rows and columns -> losing informations
Camera image as a function
- f(x,y) : position to light
- cameras cannot measure continuous space
Sub-sampling camera
- aliasing; dropping the rest of the rows and columns
- CMOS sensor have this problem
Signal to Noise Ratio (SNR)
- low light -> high SNR
- we should collect as much light as we can
- bigger sensors, larger fill factor
- minimize off time, long exposures (slower shutter speed)
3.2 Image sensors
CMOS
- (+) random access to pixels
- (+) cheaper
- (-) lower fill factor
CCD
- (+) larger fill factor
- (-) harder to read out image
- (-) more expensive
Rolling shutter
- Camera sensors work like a scanner.
- Reading values from each sensor cannot happen simultaneously.
- Fast moving object will create an image different from our sight
3.3 Image sampling
Aliasing
Nyquist rate
- is a minimum sampling rate to avoid aliasing
= two samples per cycle
Fourier transform
- decomposes any signal into components
- we can figure out max frequency in components
- to avoid aliasing, samplint rate >= 2 * max frequency
Low pass filtering (optical)
- allows the low frequencies to pass
3.4 Image scaling
Image as array of pixels
- F[x,y] : pixels
- f(x,y) : a continuous function
3.5 Filtering
- Mean filtering
- mean value from nearby 8 & 1 itself
Up-resizing
3.6 Aliasing
3.7 Format
- Compression ways
- Lossless compression
- Lossy compression
- Formats
- JPEG
- GIF
- PNG