A Novel Steerable Filter in the Frequency Domain: The Rose Curve Filter

English (OpenAlex)

Feature extraction in image processing is a difficult task. To do this, a filter bank is one of the most used techniques. To extract a feature, the image is masked with filters in prepared filter banks, one by one. If a greater number of features can be extracted from an image, the task is easier. Thus, identifying more features is desirable in many image processing applications. However, filter preparation can be troublesome because of the difficulties in determining a filter’s parameters such as direction. Limitations of the selected filter are another issue. Thus, an easy-to-use filter with few parameters and directional flexibility is a desirable choice. For these reasons, a novel steerable type of filter is proposed in this study. The proposed rose curve filter uses Lemniscate shapes derived from the rose curves to extract image features in the frequency domain. It has three parameters, fewer than other filters, and has directional flexibility. It can also extract images in more than one direction. Experimental results show that it is effective in feature extraction during image processing.