Automatic Color Matching with a Computer Kei Takahashi

  • Slides: 16
Download presentation
Automatic Color Matching with a Computer Kei Takahashi Dept. of Computer Science and Engineering

Automatic Color Matching with a Computer Kei Takahashi Dept. of Computer Science and Engineering Helsinki University of Technology

How do you like these color pairs? Universally accepted pairs

How do you like these color pairs? Universally accepted pairs

How about these? Controversial pairs (Should be avoided)

How about these? Controversial pairs (Should be avoided)

Computerized Color Matching 1) Color matching: its importance 2) Our Approach: the judge machine

Computerized Color Matching 1) Color matching: its importance 2) Our Approach: the judge machine 3) Application: automatic color adviser

1. 1 Color matching: its importance Judge if two (or more) colors match or

1. 1 Color matching: its importance Judge if two (or more) colors match or not ▌ Depend on persons and situations ▌ Still there are universally accepted pairs ▌ 1. Color matching: its importance

1. 2 Commercial importance ▌ Coloring makes all the difference n It ▌ changes

1. 2 Commercial importance ▌ Coloring makes all the difference n It ▌ changes customer’s impression Colorings improve/diminish the value n Should ▌ avoid ”controversial pairs” Needs to be unique and impressive n Always new color designs are needed 1. Color matching: its importance

1. 3 Growths of demands PC program offers infinite colors ▌ Everyone need to

1. 3 Growths of demands PC program offers infinite colors ▌ Everyone need to design ▌ Power. Point, Word, websites… Always unique design is needed Color design is a tough work n Can’t afford to put out to professional designers n 1. Color matching: its importance

2. 1 The judge machine ▌ We designed a program which judges if a

2. 1 The judge machine ▌ We designed a program which judges if a color pair is good or bad Good! Bad! 2. Our proposal : Judge machine

2. 2 Machine Learning ▌ A computer can learn general knowledge from training data

2. 2 Machine Learning ▌ A computer can learn general knowledge from training data n Not much number of training data are needed Unknown data Good examples Bad examples 2. Our proposal : Judge machine Knowledge Good!

2. 2 Machine Learning (cont. ) ▌ Prepare training data n Evaluated by “teacher”

2. 2 Machine Learning (cont. ) ▌ Prepare training data n Evaluated by “teacher” n Not so many data is needed (ex. 100) ▌ Train a computer with the data n The ▌ computer acquires general knowledge Now it can judge any color pairs n Acts just as the teacher 2. Our proposal : Judge machine

2. 3 Prototype experiment ▌ Prepared data : n Classified good pairs and bad

2. 3 Prototype experiment ▌ Prepared data : n Classified good pairs and bad pairs by hand n 200 examples ▌ Trained with the first 100 pairs Good Pairs Prepared Data Bad Pairs 2. Our proposal : Judge machine Used for training Used for Evaluation

2. 3 Prototype experiment (cont. ) Compared the judge of the computer and teacher

2. 3 Prototype experiment (cont. ) Compared the judge of the computer and teacher (human) ▌ Accuracy was 80% >> Learning was successful! ▌ Computer Teacher Good Bad 2. Our proposal : Judge machine

3. Automatic Color Adviser ▌ Generate infinite good color pairs n Generate random color

3. Automatic Color Adviser ▌ Generate infinite good color pairs n Generate random color pairs n Select pairs that passed the judgment Generate pairs randomly 3. Automatic Color Adviser Judge machine Good pairs are selected

3. Automatic Color Adviser (cont. ) ▌ Practical applications : n Website design n

3. Automatic Color Adviser (cont. ) ▌ Practical applications : n Website design n Power. Point presentations n Word documents ▌ There are no similar function for now 3. Automatic Color Adviser

Conclusion Color matching is an important area ▌ Judge machine divides good/bad color pairs

Conclusion Color matching is an important area ▌ Judge machine divides good/bad color pairs ▌ Infinite number of good color pairs are obtained in no time (Automatic Color Adviser) ▌

Further Information ▌ Our website: n http: //www. sodan. ecc. u-tokyo. ac. jp/~kei/cl/ ▌

Further Information ▌ Our website: n http: //www. sodan. ecc. u-tokyo. ac. jp/~kei/cl/ ▌ SVM (machine learning) n http: //www. csie. ntu. edu. tw/~cjlin/libsvm/