Terminator3
Advanced Member level 3
I read few papers related to Neutrosophic sets. Main idea of all papers is to transform image (or other signal) to True (T), Indeterminacy (I) and False (F) values. In case of RGB or grayscale image we must get (T,I,F) image, and then do segmentation using that resulting image. Sometimes authors propose that F=1-T, other authors use separate formula for F. From what i have read i understood, that I value is basically gradient value (first derivative). When I is bigger than alpha threshold, then pixel is with high indeterminacy. If I is small, then T and F is analyzed with some histogram-based threshold.
1. How to correctly use second derivative here? In signal processing first+second derivative is good for curve detection. Adding Gaussian blur and performing first+second derivative algorithms i can calculate first and second derivative. Edges on image will correspond to small first derivative and big value for second derivative. I imagine this values can be used as (T,F). But gradient is already used in (I) value in all papers. Anybody have ideas about why authors do not use second derivative? What value can be used for (I)?
2. What formulas can be used to generate T,I,F values to analyze audio data (waveform/spectrum)? I imagine spectrum can be analyzed as image. What about waveform data?
1. How to correctly use second derivative here? In signal processing first+second derivative is good for curve detection. Adding Gaussian blur and performing first+second derivative algorithms i can calculate first and second derivative. Edges on image will correspond to small first derivative and big value for second derivative. I imagine this values can be used as (T,F). But gradient is already used in (I) value in all papers. Anybody have ideas about why authors do not use second derivative? What value can be used for (I)?
2. What formulas can be used to generate T,I,F values to analyze audio data (waveform/spectrum)? I imagine spectrum can be analyzed as image. What about waveform data?
