<p>Now-a-days, ordinal patterns are shown to be effective in extracting discriminant image features. In this paper, we present the ordinal matrix encoding (OME) as a method that transforms an 8-bit encoded image into another image with <i>s</i> gray levels. Such an encoding acts as a highpass filter and allows us to enhance the image contours that are useful for feature extraction. In this work, we hybridized the OME technique with the linear discriminant analysis (LDA) approach to define the modified LDA (MLDA) to extract image features. The MLDA considers only interclass matrices of encoded images to highlight their singularities. Subsequently, a support vector machine (SVM) is applied to the MLDA output to perform facial image classification. We validated the proposed classification method using images from the ORL, FERET and FEI standard databases. The results indicate an overall accuracy of <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(99.07\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>99.07</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(73.61\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>73.61</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq3.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(98.78\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>98.78</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> for the ORL, FERET and FEI databases, respectively. Further, we evaluated the impact of OME by analyzing the classification accuracy of the SVM-LDA combination on raw images from the ORL database. The accuracy was <InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq4.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(95.25\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>95.25</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> with intraclass matrices and <InlineEquation ID="IEq5"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq5.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(94.50\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>94.50</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> without, both lower than the <InlineEquation ID="IEq6"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4435_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">\(99.07\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>99.07</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> achieved with encoded images. This improvement occurs because OME preserves only the essential details of the raw images for feature extraction, enhancing their discriminative ability.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Ordinal matrix encoding based facial recognition

  • Guillène Martiale Wandja,
  • J. S. Armand Eyebe Fouda,
  • Hermann Djeugoue Nzeuga,
  • Samrat L. Sabat,
  • Wolfram Koepf

摘要

Now-a-days, ordinal patterns are shown to be effective in extracting discriminant image features. In this paper, we present the ordinal matrix encoding (OME) as a method that transforms an 8-bit encoded image into another image with s gray levels. Such an encoding acts as a highpass filter and allows us to enhance the image contours that are useful for feature extraction. In this work, we hybridized the OME technique with the linear discriminant analysis (LDA) approach to define the modified LDA (MLDA) to extract image features. The MLDA considers only interclass matrices of encoded images to highlight their singularities. Subsequently, a support vector machine (SVM) is applied to the MLDA output to perform facial image classification. We validated the proposed classification method using images from the ORL, FERET and FEI standard databases. The results indicate an overall accuracy of \(99.07\%\) 99.07 % , \(73.61\%\) 73.61 % and \(98.78\%\) 98.78 % for the ORL, FERET and FEI databases, respectively. Further, we evaluated the impact of OME by analyzing the classification accuracy of the SVM-LDA combination on raw images from the ORL database. The accuracy was \(95.25\%\) 95.25 % with intraclass matrices and \(94.50\%\) 94.50 % without, both lower than the \(99.07\%\) 99.07 % achieved with encoded images. This improvement occurs because OME preserves only the essential details of the raw images for feature extraction, enhancing their discriminative ability.