SEMG signal compression based on two-dimensional techniques

Abstract

Background: Recently, two-dimensional techniques have been successfully employed for compressing surface electromyographic (SEMG) records as images, through the use of image and video encoders. Such schemes usually provide specific compressors, which are tuned for SEMG data, or employ preprocessing techniques, before the two-dimensional encoding procedure, in order to provide a suitable data organization, whose correlations can be better exploited by off-the-shelf encoders. Besides preprocessing input matrices, one may also depart from those approaches and employ an adaptive framework, which is able to directly tackle SEMG signals reassembled as images.

Methods: This paper proposes a new two-dimensional approach for SEMG signal compression, which is based on a recurrent pattern matching algorithm called multidimensional multiscale parser (MMP). The mentioned encoder was modified, in order to efficiently work with SEMG signals and exploit their inherent redundancies. Moreover, a new preprocessing technique, named as segmentation by similarity (SbS), which has the potential to enhance the exploitation of intra- and intersegment correlations, is introduced, the percentage difference sorting (PDS) algorithm is employed, with different image compressors, and results with the high efficiency video coding (HEVC), H.264/AVC, and JPEG2000 encoders are presented.

Results: Experiments were carried out with real isometric and dynamic records, acquired in laboratory. Dynamic signals compressed with H.264/AVC and HEVC, when combined with preprocessing techniques, resulted in good percent root-mean-square difference [Formula: see text] compression factor figures, for low and high compression factors, respectively. Besides, regarding isometric signals, the modified two-dimensional MMP algorithm outperformed state-of-the-art schemes, for low compression factors, the combination between SbS and HEVC proved to be competitive, for high compression factors, and JPEG2000, combined with PDS, provided good performance allied to low computational complexity, all in terms of percent root-mean-square difference [Formula: see text] compression factor.

Conclusion: The proposed schemes are effective and, specifically, the modified MMP algorithm can be considered as an interesting alternative for isometric signals, regarding traditional SEMG encoders. Besides, the approach based on off-the-shelf image encoders has the potential of fast implementation and dissemination, given that many embedded systems may already have such encoders available, in the underlying hardware/software architecture.

Keywords: HEVC; Multidimensional multiscale parser; Preprocessing technique; SEMG.

Fig. 1

Example of a one-dimensional noninvasive electromyographic (SEMG) record rearranged into a two-dimensional array

Fig. 2

Spectrum of an SEMG image

Fig. 3

Example regarding the MMP segmentation scheme, for a block of size $16\times 16$, which is repeated until the lowest scale is reached

Fig. 4

a Segmentation of an input image block and b its corresponding optimal tree

Fig. 5

Segmentation of an image block that is encoded with prediction techniques

Fig. 6

Prediction techniques (angles) used to enhance MMP

Fig. 7

Result of percentage difference sorting: a original image and b reordered image

Fig. 8

The proposed compression scheme

Fig. 9

SEMG signal retrieved from the test database

Fig. 10

Performance of the modified two-dimensional MMP encoder with isometric signals

Fig. 11

Performance of the proposed scheme regarding isometric signals, for a PDS and HEVC, b SbS and HEVC, c PDS and H.264/AVC, d SbS and H.264/AVC, e PDS and JPEG2000, and f SbS and JPEG2000

Fig. 12

Compression result for a segment from one of the test isometric SEMG signals, at a CF of 87.3 %: a original, b reconstructed after being encoded with the modified two-dimensional MMP, and c error signal

Fig. 13

Compression result for a segment from one of the test isometric SEMG signals, at a CF of 87.3 %: a signal reconstruction with SbS + HEVC, and b error signal, c signal reconstruction with PDS + HEVC and d error signal, e signal reconstruction with SbS + H.264/AVC and f error signal, g signal reconstruction with PDS + H.264/AVC, and h error signal, i signal reconstruction with SbS + JPEG2000, and j error signal, and k signal reconstruction with PDS + JPEG2000, and l error signal

Fig. 14

Performance of the modified two-dimensional MMP encoder with dynamic signals

Fig. 15

Performance of the proposed scheme regarding dynamic signals, for, a PDS and HEVC, b SbS and HEVC, c PDS and H.264/AVC, d SbS and H.264/AVC, e PDS and JPEG2000, and f SbS and JPEG2000

Fig. 16

Compression result for a segment from one of the test dynamic SEMG signals, at a CF of 87.3 %: a original, b reconstructed after being encoded with the modified two-dimensional MMP, and c error signal

Fig. 17

Compression result for a segment from one of the test dynamic SEMG signals, at a CF of 87.3 %: a signal reconstruction with SbS + HEVC and b error signal, c signal reconstruction with PDS + HEVC and d error signal, e signal reconstruction with SbS + H.264/AVC and f error signal, g signal reconstruction with PDS + H.264/AVC and h error signal, i signal reconstruction with SbS + JPEG2000 and j error signal, and k signal reconstruction with PDS + JPEG2000 and l error signal