Compact Descriptors for Visual Search for Money Recognition

Compact Descriptors for Visual Search for Money Recognition Student: Pham Tran Huong Giang Supervisor: Dr.. Compact Descriptors for Visual Search Local descriptors Compact descriptor

Compact Descriptors for Visual Search

for Money Recognition

Student: Pham Tran Huong Giang

Supervisor: Dr Le Thanh Ha

Undergraduate thesis

Outline

 Introduction

Introduction

 International standard for visual search systems

 Will be used in a great number of visual search applications

Introduction

and tourists It is hard to recognize local money in the first use when they come to a foreign country

money recognition problem

Compact descriptor for visual search

 Retrieval

Compact Descriptors for Visual Search

Local descriptors

Compact descriptor

Detect key points Select features Describe

Compress local descriptors

Encode feature’s coordinate Aggregate global descriptor

Compact Descriptors for Visual Search

 Retrieval

Descriptor database

Query image

Extract compact descriptor

Compare global descriptors Top

match list

Compact Descriptors for Visual Search

 Pairwise matching

Matching

Match local descriptors in compressed domain

Check the geometric consistency

Homography

Query image

Extract descriptor

Extract descriptor Referent images

Approach

 To evaluate:

 Collect a set of images of money as training set

 Test by another set of images (test set)

Approach

 Collecting training dataset

 Consists of the money of 6 countries: Vietnam, Laos, Cambodia, Japan, Thailand, Singapore

 Good condition of displayed money

 Without background

 Uniform distributed light

 238 images: 160 images of banknote and 78 images of coin

Approach

Input image

Extract descriptor

Match with 160 descriptors of banknote

Find and

Extract Match with 78

Evaluate by compare with 2 threshold one

of banknote and one of coin

Result

Approach

 Testing and discussion

 Test set: 6 subsets

1 100 images of banknotes (same type with images in training dataset,

taken by me)

2 50 downloaded images from the Internet (same type with training

images)

3 60 images of coins (same type with training images, taken by me)

4 25 distractor images of banknotes

5 10 distractor images of coins

6 20 distractor images without money

Testing result

 For threshold of 100 for banknote and 50 for coin

Subset Precision Number of “not found”

results

Number of false results

Discussion

 Some successful tests in subset 1

Discussion

 Some “not found” tests in subset 1

 Comment: Money in these images is folded or strong line shine through the money

Discussion

 In subset 2: 3 “not found” cases are caused by the low quality

of images

 One “false” case because of the similarity between 2 images:

Discussion

 CDVS treats well with distractor images (subset 4, 5, 6)

 Only 2 wrong cases in subset 5 because the distractor images are similar with training images

Discussion

 For coins, the recognition accuracy is hardly acceptable because of some reasons

 Coin is poor in feature

 Strongly response to light

 Whole-colorized

 Small surface

Conclusion and future work

 Conclusion:

 CDVS brings high recognition accuracy for banknote if the image

of banknote is taken in not too bad condition

 The recognition accuracy for coin is hardly acceptable

 Collect more data

 Build an mobile application for tourists to recognize all kind of

Thank you for watching