Coarse Classification via Discrete Cosine Transform and Quantization

1

Coarse Classification via Discrete Cosine

Transform and Quantization

Presented byTe-Wei

Chiang

July 18, 2005

2

Outline

Introduction Feature Extraction and

Quantization Statistical Coarse Classification Experimental Results Conclusion

3

1 Introduction

Paper documents -> Computer codes

OCR(Optical Character Recognition)

4

Requirement of Coarse Classification

high accuracy rate, high reduction capacity, and quick classification time.

5

Two subproblems in the design of classification systems

Feature extraction Classification

6

Two Philosophies of Classification

Statistical Structural

7

Advantages of the Statistical Approach

fixed-length vector, robustness for noisy patterns, less ambiguities, and easy to implement.

8

Weaknesses of the Statistical Approach

Smear or average the parameters High dimensionality

9

Goal of this Paper

The purpose of this paper is to design a general coarse classification scheme: low dependent on domain-specific

knowledge. To achieve this goal, we need:

reliable and general features, and general classification method

10

2 Feature Extraction and Quantization

Find general features that can be applied properly to most application areas, rather than the best features.

Discrete Cosine Transform (DCT) is applied to extract statistical features.

11

2.1. Discrete Cosine Transform (DCT)

The DCT coefficients F(u, v) of an N×N image represented by x(i, j) can be defined as

where

1

0

1

0

),()()(2

),(N

i

N

j

jixvuN

vuF ),2

)12(cos()

2

)12(cos(

N

vj

N

ui

.1

,021

)(otherwise

wforw

12

The DCT coefficients of the character image of “ 佛” .

13

2.2. Quantization

The 2-D DCT coefficient F(u,v) is quantized to F’(u,v) according to the following equation:

Thus, dimension of the feature vector can be reduced after quantization.

14

Illustration of the Quantization Method

15

3 Statistical Coarse Classification

The statistical classification system is operated in two modes: Training(learning) Classification(testing)

16

3.1.Proposed coarse classification

scheme

17

Illustration of Extracting the 2-D DCT Coefficients

18

3.2. Grid code transformation (GCT)

Obtain the quantized DCT coefficient qij

Transform qij to positive integer dij

Such that object Oi can be transformed to a D-digit GC.

19

3.3.Grid code sorting and elimination

Remove the redundant information.

A reduced set of candidate classes can be retrieved from the lookup table according to the GC of the test sample.

20

4 Experimental Results

In our application, the objects to be classified are handwritten characters in Chinese paleography.

Since most of the characters in these rare books were contaminated by various noises, it is a challenge to achieve a high recognition rate.

21

18600 samples (about 640 classes) are extracted from Kin-Guan ( 金剛 ) bible. Each character image was

transformed into a 48×48 bitmap. 1000 of the 186000 samples are used

for testing and the others are used for training.

22

23

5 Conclusions

This paper presents a coarse classification scheme based on DCT and quantization.

Due to the energy compacting property of DCT, the most significant features of a pattern can be extracted and quantized for the generation of the grid codes.

24

Hence the potential candidates which are similar to the test pattern can be efficiently found by searching the training patterns whose grid codes are similar to that of the test pattern.

25

Future works

Since features of different types complement one another in classification performance, by using features of different types simultaneously, classification accuracy could be further improved.