Accurate Image Retrieval Algorithm Based on Color and Texture Feature
Step 4: Display of results: after the retrieval, if there is
similarity distance between the image retrieved and the key image, the paths of image files is listed out. The images are sorted from small to big according to the similarity distances. A smaller similarity distance suggests that the similar degree between the key image and the image retrieved is bigger. If the similarity distance is 0, it means that the key image is retrieved. The user can click on the item in the list to view and compare the weighted color feature based on HSV and texture feature of each image. At the same time, the system displays 10 images with the highest similarity and reports to the user how many images have been retrieved and whether the key image has been retrieved.
III.EXPERIMENTAL RESULTS
For the algorithm proposed in this paper, Visual C++ 6 is used as the development environment, and the MFC library is adopted to verify its effectiveness. In the experiment, four groups of images, such as natural scenery, cartoon dolls, buildings and flowers, are gotten from the network library, and each group includes 100 images. Then, three methods, such as the weighted color feature based on HSV, the gray level co-occurrence matrix and the comprehensive weighted color feature based on HSV and gray level co-occurrence matrix, are used to verify the effectiveness of the algorithm. figure 2 to figure 6 show that the retrieval results of comprehensive method is much better, and the similar images in the list are more ahead.
Figure 1. The result of retrieval algorithm based on color in natural scenery
Figure 2. The result of retrieval algorithm based on texture in natural scenery
Figure 3. The result of retrieval algorithm based on color and texture in natural scenery
Figure 4. The result of retrieval algorithm based on color in cartoon dolls
Figure 5. The result of retrieval algorithm based on color in cartoon dolls
Figure 6. The result of retrieval algorithm based on color and texture in cartoon dolls
In addition, the following average precision ratio formula is also adopted to evaluate and compare various methods [20-22]: Re Re Re trieved levant precision trieved ∩ = (17)
For example, 10 cartoon dolls are selected from the image library as the key image and the top 10 images in each retrieval list are chosen, to compute the average precision and retrieval time. From Table 1, it can be found that the precision based color and texture is much higher if the average retrieval time is ignored.
TABLE 1AVERAGE PRECISION AND RETRIEVAL TIME BASED ON THREE METHODS
Retrieval algorithm Average precision (%)
Average retrieval time (s) The weighted color
feature based on HSV 55.27 0.8
The gray level
co-occurrence matrix 70.65 0.9
The comprehensive
algorithm in this paper 85.58 1.8
TABLE 2.THE SELECTION OF WEIGHT IN THE COMPREHENSIVE ALGORITHM OF THIS PAPER
Color and weight(w w1, 2) Natural scenery (%) Cartoon dolls (%) Buildings (%) Flowers (%) 0.1,0.9 62.36 59.29 51.32 71.35 0.2,0.8 64.63 60.85 57.02 72.28 0.3,0.7 67.78 63.34 72.57 75.53 0.4,0.6 78.24 72.25 63.15 85.56 0.6,0.4 68.35 83.14 57.69 74.48 0.7,0.3 63.43 59.74 51.48 69.55 0.7,0.3 61.68 58.75 50.36 68.85 0.1,0.9 60.62 57.46 50.28 67.65
From table 2, we can find that, for different weights of
color and texture, the results of the retrieval based color and texture have different effects. If unsatisfied with the retrieval result based on color or texture, the user can set the color and texture weights according to the specific situation. For example, for the image with bright color and unobvious texture, a larger color weight and a smaller texture weight can be set; But for the image, of which the texture is more obvious and the image color is not bright, a larger texture weight and a smaller color weight can be set. In a word, the retrieval based on color and texture can usually have a better effect in the process of color and texture weight adjustment.
IV.CONCLUSION
First of all, this algorithm can be used to retrieve the key image in the image library accurately. Secondly, from a large number of tests, it can be found that single color retrieval is more accordant with the visual and psychological characteristics of the character while compared with the texture retrieval, which indicates that good results can usually be obtained if the weighted color feature based on HSV is combined with the Euclidean distance. For images with obvious texture feature, the results of the texture retrieval are satisfied, but for the others, the results are not very ideal and are quite different from those of the retrievals based on color. Also, it shows that the texture feature of the gray level co-occurrence matrix is not consistent with the character’s visual characteristics, and it is necessary to combine the color feature with other features to get more ideal results. The experimental results verified the effectiveness of the method in this paper.
REFERENCES
[1] Arnold W. M. Smeulders, Senior Member, Marcel Worring. Content-Based Image Retrieval at the End of the Early Years. IEEE Transactions on pattern analysis and machine intelligence, 2000, 22(12): 1349-1380.
[2] Gong M,Fu Z.Zhang Y S.Image Retrieval Based on Integrate BTC Color Moment and GLCM . Video Engineering, 2012, 36(11): 30-33,40.
[3] Zhang G,Ma Z M.An approach of using Gabor wavelets for texture feature extraction. Journal of image and graphics, 2010, 15(2): 247-254.
[4] Chang Z, Hou Y Q,Li M L ,Cheng T, Liu L N. Image Retrieval Based on Combined Color with Te- xture Feature. Journal of Chinese Computer Systems, 2011, 32(1): 161-164
[5] Kwatra V, Essa I, Bobiek A, et a1. Texture Optimization for Example-based Synthesis// Proc Of SIGGRAPH’00 Angeles, USA: ACM Press. 2005: 795-802.
[6] Datta R, Joshi D, Li J, Wang J Z. Image retrieval: ideas, influences, and trends of the new age . ACM Computing Surveys, 2008, 40(2): 1-60.
[7] Wu J, Lu M, Liu C. SVM - feedback scheme within hybrid learning framework for image retrieval. Acta Electronica Sinica, 2010, 38(9): 2101-2106.
[8] Liu Y, Zhang D, Lu G. Region-based image retrieval with high level semantics using decision tree learning. Pattern Recognition, 2008, 41(8): 2554-2570.
[9] Wan X. Content based image retrieval using manifold-ranking of blocks [A]. Proc of IEEE Int Conf on
Multimedia and Expo (ICME'07). Beijing: IEEE Computer Society, 2007. 2182-2185.
[10] Li F, Dai Q, Xu W, Er G. Correlated probabilistic label propagation for region-based image retrieval [A]. Proc of IEEE Int Conf on Acoustics, Speech and Signal Processing (ICASSP'07). Hawaii: IEEE Computer Society, 2007: 789-792.
[11] S H Feng, D Xu, X Yang. Attention-driven salient edge(s) and region(s) extraction with application to CBIR . Signal Processing, 2010, 90: 1-15.
[12] Rahmani R, Goldman S A. MISSL: multiple- instance semi-supervised learning. Proc of Int Conf on Machine Learning (ICML'06). Pittsburgh: Association for Computing Machinery, 2006, 705-712.
[13] Wang Q,Peng J Y,Guo S.han. Approach for image retrieval based on hybrid features kernel. Computer Engineering and Applications, 2012, 48(15):168:171.
[14] Wang X Z,Wang X F.Image retrieval approach integrated color and multi-scale texture features by Adaboost algorithm. Electronic Design Engineerint, 2012, 20(4): 165-169.
[15] Song Z,Yan Y F.Image retrieval based on combined features of interest point. Journal of Computer Applications, 2012, 32(10): 2840- 284, 2847.
[16]Cinque L, Levialdi S, Pellicano A. Color-based Image Retrie Using Spatial Chromatic Histograms//Proceeding of l999 IEEE International Conference on Multimedia
Computing and Systems: IEEE Computer Society, 1999:969-973.
[17]Kwatra V, Essa I, Bobiek A, et a1. Texture Optimization for Example-based Synthesis//Proc Of SIGGRAPH’00 Angeles, USA：ACM Press．2005:795-802.
[18]Datta R ,Joshi D, Li J,Wang J Z.Image retrieval:ideas,influences,and trends of the new age. ACM Computing Surveys, 2008, 40(2): 1-60.
[19]Wang J Z,Li J,Wiederhold G. SIMPLIcity: semantics-sensitive integrated matching for picture libraries. IEEE Trans on Pattern Analysis and Machine Intelligence, 2001, 23(9):947-963.
[20]Rahmani R,Goldman S A,ZhangH,Cholleti S R,Fritts J E.Localized content based image retrieval. IEEETranson Pattern Recognition and Machine Intelligence, 2008, 30(11): 1902-1912.
[21Rahmani R, Goldman S A,Zhang H, Krettek J,Fritts J.Localized content based image retrieval[A].Proc of ACM Int Conf on Multimedia Information Retrieval (ACMMIR'05). Singapore: Association for Computing Machinery, 2005, 227-236.
[22]A. Shinya, K. Hiroaki. Content-based Image Retrieval System by Multi-dimensional Feature Vectors. International Conference on Digital Printing Technologies, 2003, 841-845.