machine learning - Ways to improve Image Pixel Classification -
machine learning - Ways to improve Image Pixel Classification -
here problem trying solve:
goal classify pixels of colored image 3 different classes. we have set of manually classified info training purposes pixels not correlate each other (each have individual behaviour) - classification on each individual pixel , based on it's individual features. 3 classes approximately can mapped colors of red, yellowish , black color families. we need have scheme semi-automatic, i.e. 3 parameters command probability of presence of 3 outcomes (for final well-tuning)having in mind:
which classification technique choose? what pixel features utilize classification (rgb, ycc, hsv, etc) ? what modification functions take well-tuning between 3 outcomes.my first seek based on
naive bayes classifier hsv (also tried rgb , ycc) (failed find proper functions well-tuning)any suggestion? thanks
for each pixel in image seek using histogram of colors n x n window around pixel features. general-purpose color matching under varied lighting conditions, have had luck using two-dimensional histograms of hue , saturation relatively little number of bins along each dimension. depending upon lighting consistency might create sense straight utilize rgb values.
as classifier, manual-tuning requirement expressed using class weights: parameters specify relative costs of false negatives versus false positives. have used functionality svms, i'm sure can find implementations of other classifiers back upwards similar concept.
image-processing machine-learning classification
Comments
Post a Comment