![]() | Only 14 pages are availabe for public view |
Abstract This chapter has presented a robust method for identification of landmines from acoustic images using SVM based on MFCC and the sine, cosine and wavelet transform techniques. Firstly, the MFCCs feature coefficients are extracted from the images after transformation to 1-D signals by three different ways (lexicographic transformation or block by block scan or spiral scan) and the DST, DCT and DWT of these signals and/ or concatenation of each with the original image signals to form a large vector called a feature vector for each image. So for all images, we have a feature matrix. Secondly, this feature matrix is used to train an SVM classifier. Experimental results have proven that the proposed method is useful for feature matching of images as a new application for MFCC technique as it is always used for speech recognition. The best performance is achieved by features extracted from the DST of images contaminated by AWGN, features extracted from the image signals plus the DCT of image signals contaminated by impulsive noise and features extracted from the image signals plus the DST of image signals contaminated by speckle noise. Also SVM takes less time than that of ANN except in the case of DCT and DST of signals. |