4. Color Feature Extraction

  • 4.1 Introduction

  • 4.2 Color Space

  • 4.3 Image Clustering and Segmentation

    • 4.3.1 K-Means Clustering
    • 4.3.2 JSEG Segmentation
  • 4.4 Color Feature Extraction

    • 4.4.1 Color Histogram

      • Description of color distribution.

      • Bin creation and pixel counting.

        • Ways to compute histograms
      • 4.4.1.1 Component Histogram

        • Splitting into individual R, G, B channels.
        • Concatenation of individual channel histograms.
        • Color quantization to reduce histogram length.
      • 4.4.1.2 Indexed Color Histogram

        • Quantizing the RGB color space into N colors (global palette).
        • Histogram creation using indexed colors as bins.
      • 4.4.1.3 Dominant Color Histogram

        • Creating the histogram from a native palatte
        • Using adaptive/native palettes created from the image itself.
        • Dominant color extraction (thresholding or K-means).
        • Advantages and limitations of histograms.
    • 4.4.2 Color Structure Descriptor

      • Addressing the lack of spatial information in histograms.
      • Using a structuring element (window) for color counting.
      • Color Structure (CS) histogram computation.
      • Multiplying effect on isolated vs. grouped colors.
      • Robustness and dependence on window size.
    • 4.4.3 Dominant Color Descriptor (DCD)

      • Based on visually interpreting images using dominant colors.
      • DCD extraction: histogram thresholding, (ci, pi) representation.
      • Region-based DCDs for spatial information.
      • Translation to color names.
    • 4.4.4 Color Coherence Vector

      • Incorporating spatial information into histograms.
      • Dividing histogram bins into coherent and non-coherent components.
      • CCV computation procedure.
      • Advantages and limitations.
    • 4.4.5 Color Correlogram

      • Color version of gray level co-occurrence matrix.
      • Characterizing the distribution of color pairs.
      • 3D histogram representation (color pairs, spatial distance).
      • Correlogram computation.
      • Autocorrelograms.
      • Matching and application
    • 4.4.6 Color Layout Descriptor

      • Capturing the frequency of color changes (similar to texture).

      • CLD computation:

        • Image partitioning (8x8 grid).
        • Color quantization (dominant color per block).
        • DCT transform.
        • Zigzag scanning and coefficient selection.
      • Scalability and limitations.

    • 4.4.7 Scalable Color Descriptor
      • Using wavelet transforms to create scalable histogram.

      • Computing steps.

      • Advantages.

    • 4.4.8 Color Moments
      • Descriptive statistics for color images.
      • Mean, variance, skewness, etc.
      • Computation from individual channels or histograms.
      • Concatenation.
      • Advantages and limitations (concise but inaccurate).
  • 4. 5 Summary

    • Importance of color space choice.
    • Role of histogram in color feature extraction.
    • Scalability and invariance properties of histograms.
    • Matching histograms (simple vs. quadratic distance).
    • Computation cost of different descriptors.
  • 4.6 Excercises

    • Calculate histogram, log transform,
    • Generate histograms, decompose images.