Various underwater studies using underwater sonar sensors are actively in progress. However, unlike the ground, the underwater has a lot of noise. So it is difficult to accurately recognize the underwater environment. The final purpose of this study is to improve the efficiency of the underwater environment recognition using the underwater sonar sensor by developing a filtering algorithm that removes noise and expresses the object from the underwater sonar image captured by the underwater sonar sensor. To develop a filtering algorithm, convolutional calculations were used with three types of filters. This paper is about the case study that conducted to set the parameters of ‘Gabor Filter’ suitable for underwater sonar image during the design process of filtering algorithm. As a result, it was possible to find the most suitable ‘Gabor Filter’ parameters for underwater sonar images. And it showed high accuracy with a binary map of obstacles created by hand using the naked eye. Through this study, it can be utilized not only as a binary map of real-time obstacles, but also as an algorithm for generating object masks in underwater sonar images for deep learning.