The SYMBIOSIS project aims to obtain images of detected fish that are classified by type. The system will be deployed in depth of 5-22m in three different marine observatories, each for a period of a month. To explore the perfromance of the system on-the-fly, the images will be transmitted from the submerged unit to the surface unit. Due to the lack of cable connection (a cable may switch and turn), the communication will be performed acoustically. Yet, since the buad rate of the underwater acoustic communication is low (roughly 10kbps) and since the size of the captured images is on the order of 10kB, there is a need to compress the image.
Our imvestigations showed that regular image compression methods fail to reduce the size of the images to a resonable level from the respect of the acoustic modem. For example, the binary black&white image shown in Figure 1 is originally of size 16kB, and is compressed to size 12kB by a lossy JPG protocol (Black & White). For this reason, we turned to develop our own image compression technique that is spesifically tailored to the case considered.
Our method builds upon the understanding that, for the objective of learing of the functionality of the system, the spesific details of the fish in the image are no important. Instead, we would like to compress the image such that the general shape of the target in the image can be observed. This resambles the problem of segmentation, where only the general location of the target is of interest. We thus turn to segmentation techniques. Our methodology is throughly described in the following publications:
A. Abu and R. Diamant, "Robust Image Denoising for Sonar Imagery", IEEE Oceans, Kobe, Japan, May. 2018 (posted on the project's database and website)
A. Abu and R. Diamant, "Enhanced Fuzzy-based Local Information Algorithm for Sonar Image Segmentation", submitted to the IEEE Transactions of Image Processing