数字图像处理小波对信号压缩汇总.ppt

Wavelet Application for Signal Compression Why do we need a transform for data compression? Generally, raw data is hard for analysis. We would like some features, either globally or locally. A transform is applied to image data to obtain further information and exploit its relations, such as frequency and redundancy. Fourier Transform It is a classical transform. Frequency information can be obtained from FT. FT: Inverse FT: Features of Fourier Transform It will transform the signal from temporal space to frequency space.(In fact we can’t obtain the whole signal in a limited time) It loses all the time information since integration is performed over all times. It is difficult to discriminate signals that have some main frequency all the time from that have the same main frequency but with different time. So signals suitable for Fourier Transform are those time invariant signals. Or those applications concern frequency only. We can’t observe and need all of the information in a signal, and only look for the useful information we expect. Short Time FT If we assume the signal is stationary in certain time slots (window), we can perform FT on this part of signal and obtain frequency information as well as time information. This is so called “Short Time Fourier Transform or Addition Window Fourier Transform ” Short Time FT A window function is introduced, which has certain width T. The transform is: For each time t’, there is an FT. We obtain a time-frequency representation of the signal. Features of STFT Unlike FT, STFT will give the information of frequency at different time intervals. How could we choose the width of window function? One can’t know what frequency components exist at what instances of times. What one can know are the time intervals in which certain band of frequencies exist. Heisenberg Uncertainty Principle The principle is, . The product of width of momentum spectrum and width of position spectrum is always greater than a minimum