Rotation invariant curvelet based image retrieval & classification via Gaussian mixture model and co-occurrence features


Engin M. A., ÇAVUŞOĞLU B.

MULTIMEDIA TOOLS AND APPLICATIONS, cilt.78, sa.6, ss.6581-6605, 2019 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 78 Sayı: 6
  • Basım Tarihi: 2019
  • Doi Numarası: 10.1007/s11042-018-6368-8
  • Dergi Adı: MULTIMEDIA TOOLS AND APPLICATIONS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Sayfa Sayıları: ss.6581-6605
  • Anahtar Kelimeler: Curvelet transform, Image retrieval, Image classification, TEXTURE, TRANSFORM
  • Atatürk Üniversitesi Adresli: Evet

Özet

Demand for better retrieval methods continue to outstrip the capabilities of available technologies despite the rapid growth of new feature extraction techniques. Extracting discriminatory features that contain texture specific information are of crucial importance in image indexing. This paper presents a novel rotation invariant texture representation model based on the multi-resolution curvelet transform via co-occurrence and Gaussian mixture features for image retrieval and classification. To extract these features, curvelet transform is applied and the coefficients are obtained at each scale and orientation. The Gaussian mixture model (GMM) features are computed from each of the sub bands and co-occurrence features are computed for only specific sub band. Rotation invariance is provided by applying cycle-shift around the GMM features. The proposed method is evaluated on well-known databases such as Brodatz, Outex_TC_00010, Outex_TC_00012, Outex_TC_00012horizon, Outex_TC_00012tl84, Vistex and KTH-TIPS. When the feature vector is analyzed in terms of its size, it is observed that its dimension is smaller than that of the existing rotation-invariant variants and it has a very good performance. Simulation results show a good performance achieved by combining different techniques with the curvelet transform. Proposed method results in high degree of success rate in classification and in precision-recall value for retrieval.