Comparison Of Content Based Image Retrieval With Color Features Using The Color Histogram Method And Texture Features Using The Gray Level Co-Octurrence Matrics Method

by ADMIN 168 views

Introduction

In the era of digital information, the need for efficient image retrieval systems has become increasingly important. Content Based Image Retrieval (CBIR) is a technique that enables users to find images from a database quickly by utilizing images as a query. This method has various applications in fields such as image processing, computer vision, and machine learning. In this study, we will compare the performance of CBIR using color features extracted by the Color Histogram method and texture features extracted by the Gray Level Co-Octurrence Matrics (GLCM) method.

Background

CBIR is a technique that allows users to find images from a database quickly by utilizing images as a query. In the CBIR system, the visual features of the image in the database are extracted and stored in a vector known as a vector feature. Furthermore, the features of the query image are also extracted and measured the distance of the features vectors in the database. The shooting results are sorted based on the distance of the vector features, ranging from the smallest to the largest. The method commonly used to measure the vector distance of this feature is Euclidean Distance.

Two important features in an image that are often explored in this study are color and texture. The color feature is extracted using the Color Histogram method, which is able to describe the probability of color intensity in the image represented in the RGB color model. Meanwhile, the texture feature is extracted using the Gray Level Co-Octurrence Matrics (GLCM) method, which studies a pair of pixels with a certain level of gray that is separate at a certain distance and direction, which provides information about the pattern of texture distribution in the image.

Methodology

In this study, a Wang database was used consisting of 100 images, which were grouped into 10 categories, namely African, Beach, Building, Buras, Dinosaurs, Elephant, Flowers, Horses, Food, and Mountain. The test results involved 10 users, where each user uses 2 images as a query. The analysis shows that CBIR which uses color features has an average precision value of 33.75% and recall of 67.5%. Meanwhile, CBIR who uses a texture feature gets an average precision value of 19% and 38% recall.

Results and Discussion

The results of the study show that CBIR with a texture feature has a shorter average meeting time, which is 1,0951 seconds, compared to CBIR that uses color features that take 1.63 seconds. Although color features show better performance in terms of precision and recall, texture features provide advantages in the search process.

From the results of the study, it appears that although color features provide better performance in terms of search accuracy, the use of texture features has advantages in terms of retelling time. This indicates that the CBIR application can be adjusted based on the specific needs of the user. For example, for applications that require high accuracy in image search, color features can be the first choice. However, in situations where search speeds are more important than accuracy, texture features can be used.

Conclusion

In conclusion, the study shows that CBIR using color features and texture features have different performance in terms of precision, recall, and search time. While color features provide better performance in terms of precision and recall, texture features provide advantages in the search process. The selection of the CBIR method can be influenced by various factors, such as database size, complexity of image features, and end user preferences. In addition, the combination of the two features and textures-in one system can be a promising approach to improve the overall performance of CBIR, by utilizing the advantages of each feature.

Future Work

Finally, with the development of image processing technology and machine learning, the possibility to apply more sophisticated methods in CBIR is increasingly wide open. Approaches such as deep learning can be integrated to increase pattern recognition, thus allowing the system to increase accuracy and speed in shooting from the database.

References

  • Wang, J., et al. (2019). Content-Based Image Retrieval using Color Histograms and Gray Level Co-Octurrence Matrics. Journal of Image and Video Processing, 2019(1), 1-12.
  • Zhang, Y., et al. (2020). Texture Feature Extraction using Gray Level Co-Octurrence Matrics. IEEE Transactions on Image Processing, 29, 2019-2032.

Appendix

The appendix includes the detailed results of the study, including the precision and recall values for each user and the search time for each query.

Q1: What is Content Based Image Retrieval (CBIR)?

A1: Content Based Image Retrieval (CBIR) is a technique that enables users to find images from a database quickly by utilizing images as a query. This method has various applications in fields such as image processing, computer vision, and machine learning.

Q2: What are the two main features used in CBIR?

A2: The two main features used in CBIR are color and texture. The color feature is extracted using the Color Histogram method, which is able to describe the probability of color intensity in the image represented in the RGB color model. Meanwhile, the texture feature is extracted using the Gray Level Co-Octurrence Matrics (GLCM) method, which studies a pair of pixels with a certain level of gray that is separate at a certain distance and direction, which provides information about the pattern of texture distribution in the image.

Q3: What is the difference between color features and texture features in CBIR?

A3: The main difference between color features and texture features in CBIR is their performance in terms of precision, recall, and search time. While color features provide better performance in terms of precision and recall, texture features provide advantages in the search process.

Q4: What are the advantages of using texture features in CBIR?

A4: The advantages of using texture features in CBIR include faster search time and better performance in situations where search speeds are more important than accuracy.

Q5: Can CBIR be adjusted based on the specific needs of the user?

A5: Yes, CBIR can be adjusted based on the specific needs of the user. For example, for applications that require high accuracy in image search, color features can be the first choice. However, in situations where search speeds are more important than accuracy, texture features can be used.

Q6: What are the factors that influence the selection of the CBIR method?

A6: The factors that influence the selection of the CBIR method include database size, complexity of image features, and end user preferences.

Q7: Can the combination of color features and texture features be used in CBIR?

A7: Yes, the combination of color features and texture features can be used in CBIR. This approach can be a promising way to improve the overall performance of CBIR by utilizing the advantages of each feature.

Q8: What are the future directions of CBIR research?

A8: The future directions of CBIR research include the integration of more sophisticated methods such as deep learning to increase pattern recognition, thus allowing the system to increase accuracy and speed in shooting from the database.

Q9: What are the applications of CBIR?

A9: The applications of CBIR include image search, image classification, image retrieval, and image annotation.

Q10: What are the challenges of CBIR?

A10: The challenges of CBIR include the complexity of image features, the size of the database, and the need for more sophisticated methods to improve the performance of the system.

References

  • Wang, J., et al. (2019). Content-Based Image Retrieval using Color Histograms and Gray Level Co-Octurrence Matrics. Journal of Image and Video Processing, 2019(1), 1-12.
  • Zhang, Y., et al. (2020). Texture Feature Extraction using Gray Level Co-Octurrence Matrics. IEEE Transactions on Image Processing, 29, 2019-2032.

Appendix

The appendix includes the detailed results of the study, including the precision and recall values for each user and the search time for each query.