Multi-resolution Fourier-based texture image retrieval

During the last decades, several approaches were proposed to describe texture contents of an image. In early research works, the texture features were mainly extracted from the pixel space itself (Edge histograms, Co-occurrence-based features). Later on, the focus was more on the use of dual spaces (transform of pixel space) such as frequency space or spaces resulting from Gabor and wavelet transforms. Recent physiological studies showed that the human visual system can be modeled as a set of independent channels of various orientations and scales, this finding motivated the proliferation of multi-resolution methods for describing texture images. Most of these methods are either wavelet-based or Gabor-based. In this paper we propose a multi-resolution technique for characterizing and retrieving images which works on the Fourier domain. Our approach differs from the work of [3] since it applies global Fourier transform on a hierarchy of images of various resolutions while their approach uses local (windowed) Fourier transform of increasing sizes. The experiments we have conducted showed that the multi-resolution approach improves the retrieval accuracy of the similar method that uses the same texture feature vector but in a single resolution. The experiments have also demonstrated that our Fourier based multi-resolution technique outperforms many wavelet-based multi-resolution techniques recently described in literature.