I've been in crossroads between signal resizing before segmentation using sliding window (or) feature resizing just after computing it within the window. Because I felt that signal resizing using interpolation techniques (or) padding techniques does change the pattern of the signal. Experts help needed.
For eg. I've 2 signals with length 3500x1, 2500x1 respectively. If I extract mean feature from the 1st signal I've 25 mean features from a window length of 100(1 sec), but I've 35 features for the next signal. So my question is whether there is any alternate way of doing this. I tried a crazy idea by taking the mean of the 1ms from all the 24 window, thereby having 100 features. In this way I'm able to get a fixed feature length of 100 as opposed to 24 in the previous case. for different length signals, but an expert told it was wrong. More input needed.
As far as I understand - feature extraction itself is aimed at reduction of dimension . So resizing is explicitly not needed .
Next , when you analyze features you are no longer in the raw data domain for having a one-to-one correlation .You are in the feature space and your interpretation is based on them (not on raw data ) .
If you resize by taking the mean, you will reduce the dimensionality but also lose high-frequency details. If your feature extraction afterwards is based on low-frequency information in the signal, then losing the high frequency details is not an issue. However, if your feature extraction is based on high frequency, then you have lost a significant amount of high frequency information using your 100-tap mean filter.
If your aim is to fix the feature vector size, you can look at techniques that will capture the essence of the information in your data and have a fixed feature vector size. Consider for example, the pdf of the data.
whenever you go for interpolation and padding mean value changes from the actual mean. It is better to find out the solution without interpolation and padding
No need to resize your signal before segmentation part.Yes obviously any signal resizing technique change the pattern of the signal...It leads to change in orientation of data points in the signal.
hi
A new image-re-sizing method using seam carving and a saliency Strength Map . Re sizing would be an ideal procedure to handle with the different sampling rate from the Before the segmentation and feature extraction process, the ECG signals were prepossessed.
Dear Mary mam,
Does resizing overcome the difference in signal lengths (or) duration of signals?
Before the segmentation and feature extraction process of any digital image or signals re-sizing process must need to analyze the next process to implement the common procedure
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts