PaperlayerView exact revision on arXiv

Source-linked AI summary

Comparing the Performance of L*A*B* and HSV Color Spaces with Respect to Color Image Segmentation

Dibya Jyoti Bora, Anil Kumar Gupta, Fayaz Ahmad Khan

arXiv:1506.01472v1cs.CV

TL;DR

Color image segmentation requires choosing an appropriate color space, with L*A*B* and HSV frequently used. The paper compares these spaces using a segmentation approach and evaluates them with MSE and PSNR; HSV is reported to perform better than L*A*B*.

  • Problem

    Choosing an appropriate color space is an important issue in color image segmentation, where L*A*B* and HSV are frequently chosen.

  • Method

    The paper compares L*A*B* and HSV using a color image segmentation approach with clustering and evaluates performance using MSE and PSNR.

  • Results

    HSV shows higher PSNR values and lower MSE values than L*A*B*, indicating better segmentation performance in the reported experiments.

  • Takeaways & Limitations

    Within the reported comparison, HSV is the better-performing color space for the segmentation task.

Abstract

from arXiv · show

Color image segmentation is a very emerging topic for image processing research. Since it has the ability to present the result in a way that is much more close to the human yes perceive, so todays more research is going on this area. Choosing a proper color space is a very important issue for color image segmentation process. Generally LAB and HSV are the two frequently chosen color spaces. In this paper a comparative analysis is performed between these two color spaces with respect to color image segmentation. For measuring their performance, we consider the parameters: mse and psnr . It is found that HSV color space is performing better than LAB.

I. INTRODUCTION

Color image segmentation depends strongly on selecting an appropriate color space. The paper compares frequently chosen L*A*B* and HSV spaces, while prior work applies varied segmentation methods in both spaces.

  • Color image segmentation can improve image analysis, but choosing a proper color space is a central issue.
  • L*A*B* and HSV are among the two frequently chosen color spaces for color image segmentation.
  • Clustering is described as an efficient methodology for image segmentation.
  • Prior HSV methods combine color-component analysis with thresholding, K-Means, mean shift, FELICM, or histogram-based clustering.
  • Prior L*A*B* methods use channel selection, genetic algorithms, K-Means, Sobel filtering, watershed processing, or clustering.

III. COLOR IMAGE SEGMENTATION

Color image segmentation partitions image content into meaningful regions, and color spaces organize color information for representation and analysis. The section introduces L*A*B* and HSV representations used in the study.

  • Color image segmentation extracts or partitions image regions and supports applications including multimedia, text extraction, and medical-image feature identification.
  • A color space organizes colors to support reproducible digital and analog color representation.
  • Different color spaces represent color information differently, making particular calculations and color identification more convenient.
  • L*A*B* Color Space: L*A*B* represents lightness with L* and chromaticity with a* and b* channels.
  • HSV Color Space: HSV represents color using Hue, Saturation, and an intensity-related vertical axis in a three-dimensional hexacone.

V. FLOWCHART OF THE APPROACH FOLLOWED

The paper adopts a previously proposed color image segmentation approach for comparing L*A*B* and HSV color spaces. A flowchart presents the resulting procedure.

  • The comparative study adopts an approach previously proposed by the authors for color image segmentation.
  • The adopted segmentation procedure is presented through a flowchart.

VI. CLUSTERING & K-MEANS ALGORITHM

The paper uses hard clustering and selects K-Means for segmentation. K-Means iteratively assigns points to clusters and updates centers, while cosine distance is chosen for its orientation-based similarity.

  • Clustering divides data elements into groups with high within-cluster similarity and low between-cluster similarity.
  • The study selects hard clustering because soft clustering requires more complex fuzzy computations.
  • K-Means Algorithm: K-Means repeatedly assigns patterns to the nearest cluster center and recomputes centers until memberships or centroids stabilize.
  • K-Means Algorithm: K-Means uses a squared-error objective function and has low computational complexity.
  • K-Means Algorithm: The number of clusters must be selected beforehand, and an incorrect choice may produce an incorrect clustering result.

VII. COSINE DISTANCE METRIC FOR K-MEANS ALGORITHM

The paper selects cosine distance for K-Means because it emphasizes vector orientation, which is considered more relevant than magnitude when comparing image pixels.

  • Cosine distance measures similarity between n-dimensional vectors by calculating the cosine of the angle between them.
  • The metric emphasizes data-point orientation rather than magnitude, motivating its selection for image-pixel clustering.

VIII. SOBEL OPERATOR

The Sobel operator detects image edges by estimating intensity gradients in horizontal and vertical directions. Its convolutional filters emphasize transitions while remaining relatively inexpensive computationally.

  • The Sobel operator emphasizes image edges and transitions by approximating the gradient of image intensity.
  • It computes gradients using two kernels: Gx estimates the x-direction gradient, while Gy estimates the y-direction gradient.
  • The operator convolves the image with small, separable, integer-valued filters in horizontal and vertical directions.
  • The absolute gradient magnitude is introduced after computing directional gradients and may be approximated for processing.
  • Sobel filtering can smooth random-noise effects and enhance edge elements, producing bright, thick edges.

IX. MARKER-CONTROLLED WATERSHED SEGMENTATION

Marker-controlled watershed segmentation uses foreground and background markers to constrain the segmentation function before computing its watershed transform. The surrounding pipeline also uses median filtering and MSE/PSNR measures for noise handling and evaluation.

  • The watershed technique is a region-based segmentation method grounded in the geographical concept of watersheds.
  • The method computes a segmentation function whose dark regions represent the objects targeted for segmentation.
  • Foreground markers are connected pixel blobs within objects, while background markers are pixels outside the objects.
  • The segmentation function is modified so that its minima occur only at the foreground and background markers before the watershed transform is computed.
  • Median filtering removes image noise while preserving edges, and MSE and PSNR quantify error and signal-to-noise quality.

XII. EXPERIMENTS & RESULTS

The experiments compare L*A*B* and HSV color spaces within a Matlab K-Means segmentation pipeline applied to the peppers demo image. Using MSE and PSNR, the reported results favor HSV over L*A*B*.

  • Experimental setup: The study uses Matlab and evaluates segmentation on the peppers image from the Matlab demo collection.
  • Experimental setup: K-Means is configured with k = 3 clusters and the cosine distance metric.
  • L*A*B* results: The L*A*B* pipeline includes color conversion, K-Means clustering, Sobel filtering, watershed processing, and median filtering.
  • HSV results: The HSV pipeline includes HSV conversion, K-Means clustering, Sobel filtering, watershed processing, and median filtering.
  • Quantitative comparison: The reported L*A*B* PSNR values are -15.5844, -7.9527, and 3.2115 for channels 1, 2, and 3, respectively.
  • Quantitative comparison: HSV produces higher PSNR values and lower MSE values than L*A*B*, which the paper interprets as better segmentation performance.

XIII. CONCLUSION

Color image segmentation is increasingly emphasized because it can produce results closer to human visual perception than grayscale segmentation. The paper compares L*A*B* and HSV color spaces using MSE and PSNR, reporting better performance for HSV.

  • Color image segmentation is more reliable than grayscale segmentation and can produce results closer to human visual perception.
  • Color space is an important segmentation parameter because the selected space should help limit noise in the segmented image.The paper identifies L*A*B* and HSV as frequently chosen color spaces for this task.
  • HSV color space performs better than L*A*B* color space according to comparisons using MSE and PSNR.

XIV. FUTURE ENHANCEMENT

Future work will study HSV color space in greater depth, including how hue, saturation, and intensity variations affect visual perception in image pixels. This analysis is intended to support optimization of color image segmentation.

  • Future work will examine HSV color space features in depth, including hue, saturation, and intensity variations for color-image pixels.
  • The planned analysis will consider visual perception when studying variations in hue, saturation, and intensity values.
  • Understanding HSV color space more fully is intended to help optimize the color image segmentation process.
Loading 1506.01472v1…