Compression of surface texture acceleration signal based on spectrum characteristics

The purpose of this study is to find a stable spectral feature and illustrate the usability of this feature in compressing the original signal. The compression method is suitable for vibrotactile acceleration and not for force and velocity data. Because the power of the force and velocity signals is concentrated in a very lowfrequency range, it is difficult to distinguish whether the trend is consistent. In addition, the method based on STFT spectral characteristics can further reduce the compression ratio for long-term data compression and the application of multi-interaction points. The proposed coding and decoding scheme is based on the frequency domain. The original waveform can be reconstructed by obtaining the normalized amplitude spectrum, the maximum value of the STFT amplitude spectrum in each time window, and a small amount of the initial phase. For the surface texture of the same material, although the amount of

data increases significantly with a longer time for data collection, or with an increase in the number of collection points, the increase in data after compression remains small; that is, the compression ratio becomes smaller. This is because, regardless of how the data increases, the normalized amplitude spectrum characteristics change little, and the data increase after compression is small. Finally, although most of the surface textures of materials we encounter in daily life are approximately isotropic and homogeneous, we must discuss the applicability of the proposed solution in the case of surface anisotropy or unevenness:

• If the actual ratio of change per unit period is less than a certain critical value, the surface is still considered to be uniform and isotropic. The short-term amplitude spectrum trend of the tactile data remained similar after filtering.

• The surface is uniform and isotropic in a small time window, or the change in the surface texture is not drastic, regardless of whether it can be perceived, and the method in this study is still applicable. The STFT normalized spectrum characteristics are compressed by the JPEG standard, which can still achieve a high compression ratio for images that change smoothly.

• In particular, if the texture of the two materials is spliced, the advantage of frequency-domain compression is proved. If the texture spectrum characteristics of the two materials are similar, no processing is required. Otherwise, it is easy to find the dividing point of the two different spectrum distributions through the normalized spectrogram, and then compress them separately.

In addition, considering this study is based on the STFT, it is necessary to treat the data of a time window as a frame. There is no problem with data compression, but if real-time compression is required, that is, the length of the data frame is reduced, the stability of this spectral feature will decrease, and the compression efficiency will also deteriorate. This is the reason why the 500ms data were previously selected for processing.

    Authors found that the STFT spectrum of the same kind of surface texture vibrotactile acceleration has a high degree of similarity, and used this similarity to encode the vibrotactile acceleration with the public database to achieve a 10% compression. Meanwhile, the modified G&L algorithm used to reconstruct the original signal achieved at least 90% time-frequency similarity. The subjective score for the similarity between the original signal and the restored signal was found to be high, with an average of 87.85. For multi-interaction points in space, the trend similarity grayscale image can be reused, and the compression ratio is further reduced.