Hi, Dear lovely researchers.. Currently, I have read some journal about feature extraction using wavelet transform. Unfortunately, it not state clearly how to plot feature vector from 2D DWT image. I'm sorry if my question too basic but I'm really appreciate those who can help my understanding.. TQ :)
Conference Paper Feature extraction using wavelet transform for neural networ...
DWT can be perceived as two filters - hi-pass and low-pass. In 1D you take the hi-pass output window, calculate its sum of squares and this way you get the single number which is the "feature". (Optionally you take the low pass output window and stream it to another set of filters; so you get the second feature, etc.).
In 2D DWT you send each row of image to two filters, so you get hi-pass and low-pass output windows for all rows again. BUT you do not convert each hi-pass content to a feature. INSTEAD you put all such hi-pass filtered rows of image to a matrix. And you apply DWT again to all colons of this matrix. Then you can calculate the first 2D feature, etc.
Disclaimer - this is off-the-shelf explanation meant for help in understanding the principles and background. In the commonly recognized convention some details and schedule can be different (inverted).
DWT can be perceived as two filters - hi-pass and low-pass. In 1D you take the hi-pass output window, calculate its sum of squares and this way you get the single number which is the "feature". (Optionally you take the low pass output window and stream it to another set of filters; so you get the second feature, etc.).
In 2D DWT you send each row of image to two filters, so you get hi-pass and low-pass output windows for all rows again. BUT you do not convert each hi-pass content to a feature. INSTEAD you put all such hi-pass filtered rows of image to a matrix. And you apply DWT again to all colons of this matrix. Then you can calculate the first 2D feature, etc.
Disclaimer - this is off-the-shelf explanation meant for help in understanding the principles and background. In the commonly recognized convention some details and schedule can be different (inverted).
It depends on the feature extracted and the area in which you want to use it?
You can show DWT coefficients (details and approximation) like a gray image. Of course you have to normalize the coefficient values to the range [0,255] before showing them.
in MATLAB Imagesc will display it as if it was an image, and it scales the "pixel" values so that they are displayed "correctly"
You can use mixed scales representation to plot the DWT coefficients
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing