Your SlideShare is downloading. ×

Thesis writing - week9

164

Published on

s1160123 …

s1160123
Tomoyuki Soeta

Published in: Technology, Design
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
164
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Information Retrieved - Image based search s1160123 Tomoyuki Soeta Supervised by Prof. Qiangfu Zhao System Intelligence Lab 1
  • 2. Outline Introduction Information Retrieval VQ (Vector Quantization) Divide into the 8x8 block Making of Code book K-means algorithm Extract each image’s feature vector Result  Image and feature vector  Distance of feature vector Conclusion Future work 2
  • 3. Introduction I want to aim at the improvement of information retrieval system to search it even if the input data are documents or images. I have charge of a research on information retrieval based on a image. To search images using a search engine, we may use the index attached to the image, the file name, etc. as the key-words. We may also use "the contents of an image themselves." I study a new image search technique based on the code book information. 3
  • 4. Information Retrieved Image Text Divide into the block (1 block 8x8) Morphological Analysis Code bookWord Filtering Code of each block Feature Vector Feature Vector NNTree or SVM 4
  • 5. VQ (Vector Quantization)  Compression coding of images  Image compression technologyimage  In my study, I use VQ to Vector translate an image into a bag- Quantization of-blocks (BOB) (VQ) same way as document search 5 feature vector
  • 6. Divide into the 8x8 block (1) I used 10 facial images with the size 256x256. images are converted to gray scale images. Divided into the block (one-block 8x8 size). Each image obtains the block of 32×32 pieces severally. 32 blocks 32 b l 1 block 8x8 o c k s 6
  • 7. Divide into the 8x8 block (2)  Block’s pixel value is read.  Pixel read value is stored in the array of 1x64.  One image can be divided into 1024 blocks, and an array of 1024 rows can be obtained. 1x64 2 3 4 6 8 3 7 2 2 3 4 6 8 3 7 2 8 2 8 2 8 2 8 2 8 21 block 8x8 ・ ・ ・ ・ 8x8 1024 rows 7
  • 8. 10 imagesImage 0 Image 1 Image 2 Image 3 Image 4Image 5 Image 6 Image 7 Image 8 Image 9 8
  • 9. Making of Code book The array of 10240 that can be done by reading 10 images is made The code book is made by using the k-means method. Making Code book (size 256) 9
  • 10. K-means algorithmStep 1) k initial "means" are randomly selected fromthe data set .Step 2) k clusters are created by associating everyobservation with the nearest mean.Step 3) The centroid of each of the k clustersbecomes the new means.Step 4) Steps 2 and 3 are repeated untilconvergence has been reached.Step 1 Step 2 Step 3 Step 4 10
  • 11. Extract each image’s feature vector (1)  The feature vector are extracted by using code book.  There is arrangement 1024 per one image.  Arranging an individual distance of the array each one and code book is measured  The number of the nearest code is returned.  Which code how many times came out is preserved as an array. 5 1x64 42 3 4 7 8 9 2 ########## Code 7 3 Code 38 Code 72 2 Code 200 1 Code 7 Code 200 Code 72 Code 38 7 0 ・ 1 256 ・ 7 38 72 200 ・ 1024 rows Code book 11
  • 12. Result – image and feature vector(1) Image 0 Image 1 12
  • 13. Result – image and feature vector(2) Image 2 Image 3 13
  • 14. Result – image and feature vector(3) Image 4 Image 5 14
  • 15. Result – image and feature vector(4) Image 6 Image 7 15
  • 16. Result – image and feature vector(5) Image 8 Image 9 16
  • 17. Result - Distance of feature vector(1)  Euclidean distance between feature vectors is measured, and the accuracy of the code book is seen. P and Q are assumed to be two feature vectors. Data : x = (x1, x2, ..., xn) and y = (y1, y2, ..., yn) n : size of the feature vector The distance of P and Q is below. 17
  • 18. Result - Distance of feature vector(2) 256 feature0 feature1 feature2 feature3 feature4 feature5 feature6 feature7 feature8 feature9feature0 0 0.279945 0.280761 0.226158 0.291376 0.322875 0.300502 0.2307 0.23509 0.228708feature1 0.279945 0 0.19849 0.271927 0.318353 0.352126 0.324807 0.272823 0.269333 0.30847feature2 0.280761 0.19849 0 0.308124 0.352732 0.378846 0.359333 0.310492 0.316141 0.324054feature3 0.226158 0.271927 0.308124 0 0.221109 0.276269 0.240734 0.09959 0.086469 0.136439feature4 0.291376 0.318353 0.352732 0.221109 0 0.222279 0.17478 0.202749 0.210865 0.248531feature5 0.322875 0.352126 0.378846 0.276269 0.222279 0 0.084866 0.282603 0.270858 0.306136feature6 0.300502 0.324807 0.359333 0.240734 0.17478 0.084866 0 0.245255 0.232873 0.276931feature7 0.2307 0.272823 0.310492 0.09959 0.202749 0.282603 0.245255 0 0.105974 0.155957feature8 0.23509 0.269333 0.316141 0.086469 0.210865 0.270858 0.232873 0.105974 0 0.152093feature9 0.228708 0.30847 0.324054 0.136439 0.248531 0.306136 0.276931 0.155957 0.152093 0 :minimum distance  The image5 and image6 is the same persons, image5 doesnt wear glasses, and image6 wears glasses.  Between feature5 and feature6 is minimum distance.Image 5 Image 6 18
  • 19. Conclusion In my research, I study a new image search technique based on the code book information. The code book is obtained using the VQ method. It is thought that an accurate feature vector was able to be extracted about the accuracy of the feature vector because the distance between Feature5 and 6 was short. Information retrieval based on "the contents of a image themselves." 19
  • 20. Future work The background is nullified. The feature vector is extracted in the block of a different size like the block of not the block of 8x8 size but 16x16 size etc. Multimedia retrieval that uses SVM. 20
  • 21. Thank you for your attention! 21

×