秦羽捷 ( Yujie QIN )'s Blog » Academic

Homework 6. of Machine Intelligence 2 – ICA

Yujie QIN — Sat, 19 Jun 2010 22:25:14 +0000

I take part in a lesson of Machine Intelligence 2 in this semester. And here is my version of the solution of the 6. exercise.

Main Taks: Use ICA to separate two mixed audio clips.

: Homework6 of Machine Intelligence 2 – ICA

My document(PDF format) of this task is below, if you have interest, maybe you can read it.

Homework 6 of Machine Intelligence 2 – ICA

Homework 5. of Machine Intelligence 1 – NN-KNN-PW-RBF

Yujie QIN — Thu, 18 Mar 2010 10:26:09 +0000

I take part in a lesson of Machine Intelligence 1 in this semester. And here is my version of the solution of the 5. exercise.

Task1: Using Neural Network with Gradient Decent, Line Search and Conjungate Gradient methods

Task2: Using KNN in classification

Task3: Using Parzen Window in classification

Task4: Using RBF in classification

My document(PDF format) of this task is below, if you have interest, maybe you can read it.

Homework 5. of Machine Intelligence 1 – NN-KNN-PW-RBF

Homework 3. of Automatic Image Analysis – Extrema and Ridge Detection

Yujie QIN — Fri, 12 Mar 2010 21:42:43 +0000

I take part in a lesson of Automatic Image Analysis in this semester. And here is my version of the solution of the 3rd. exercise.

The main task is to detectExtrema and Ridge from images.

My document(PDF format) of this task is below, if you have interest, maybe you can read it.

Homework 3. of Automatic Image Analysis – Extrema and Ridge Detection

Homework 5. of Photogrammetric Computer Vision – 3D Reconstruction

Yujie QIN — Thu, 21 Jan 2010 19:57:07 +0000

I’m also take one lesson on Computer Vision in this semester. The homeworks are not difficult but practical.

Here is output of the 5th. homework.

Figure 1.

Figure 2.

The main task is to do both Projective Reconstruction and Euclidean Reconstruction given corresponding point pairs and some control points in euclidean coordinates.

In Projective Reconstruction we firstly calculate the fundamental matrix from the given point pairs and then derivate the projective matrices using normalized configuration (the most simple way) . With all this information a linear triangulation method is used to calculate the corresponding object points in projective space. And those procedure is so called Projective Reconstruction (of course the simple without any optimization).

In Euclidean Reconstruction we simply using all the information we got from the Projective Reconstruction and make the same procedure to the control points. At the mean time we calculate the spatial homograph using the control points. Finally all the object points in projective space are transposed into euclidean space using the same spatial homograph, and that is so called Euclidean Reconstruction (without any optimization).

My document(PDF format) of this homework is below, if you have interest, maybe you can read it.

Homework 5. of Photogrammetric Computer Vision – 3D Reconstruction

Homework 2. of Automatic Image Analysis – Hough Transform

Yujie QIN — Fri, 18 Dec 2009 01:25:24 +0000

I take part in a lesson of Automatic Image Analysis in this semester. And here is my version of the solution of the second exercise.

Figure 1.

Figure 2.

Figure 3.

The task of figure 1. is to identify the origin shape after some scaling, moving, rotating operations with the help of so called Hough Transform. The task of figure 2. is a advanced version of the task before, which will identify the money from a complicated scene. The task of figure 3. is mostly used in camera calibration to automatically find out the control points, which then can be used to calculate the calibration matrix and the parameters of camera.

My document(PDF format) of this task is below, if you have interest, maybe you can read it.

Homework 2. of Automatic Image Analysis – Hough Transform

Homework 1. of Automatic Image Analysis – Fourier Descriptor

Yujie QIN — Wed, 04 Nov 2009 04:10:12 +0000

I take part in a lesson of Automatic Image Analysis in this semester. And the first exercise is really interesting, so I decide to paste it here.

Figure 1.

The two images in the left show two type of leaves that we want to identify in the right-up image, and the right-down image is the result of my algorithm.

The algorithm just need a so called Fourier Descriptor to finish the job and actually some morphology operators to find out the edges of each leaf.

My document(PDF format) of this task is below, if you have interest, maybe you can read it.

Homework 1. of Automatic Image Analysis – Fourier Descriptor

Abstract of my bachelor thesis

Yujie QIN — Thu, 24 May 2007 18:55:03 +0000

Off-line Palm Print Image Compression

using Wavelet Transform

Abstract

With the development of Bioinformatics,especially the palm print recognition technology, the requirement of a specify method for palm print image compression grows. To meet the increasingly severe social public security situation, the Palm print auto-recognitions system has been proposed as a key project in the Ministry of Public Security “11th Five-Year” plan. Palm print images with high resolution characteristics occupies with more storage space, which cause great inconvenience to data storage and data transmission of the actual system. Currently, palm print image compression specialized field of study is rarely approaches; many applications of palm print compression directly used the fingerprint image compression algorithm. However, as the basic distinction between palm print image and fingerprint images, those applications cannot achieve satisfactory results. Wavelet transform is one of the most important improvements base on the short time Fourier transform. And the data compression is a region growing with the wavelet technology. This paper focus on establishing a specify palm print compression method based on the characters of palm prints and building up the experimental system.

Refer to the JPEG methods, WSQ method and other wavelet compression methods, this paper process a sufficient new method based on the EZW method. It introduces a new DC level shifting scheme and set the lowest frequency coefficients to zero in EZW method, including pre-processing, wavelet transform and improved EZW method.

The experimental sample database includes 100 standard palm print images, and the evaluation sample database includes 15 images random picked up from the experimental sample database. By the PSNR test of whole image and print region, this paper’s method got higher PSNR than WSQ, JPEG and EZW methods in 10 time compression ratio. The system firstly built in MATLAB environment, then, as the final resolution, recode in C to a DLL library with great interface and efficiency.

Key Words

Biometrics identification, Offline Palm print Recognition, Wavelet Transform, image compression