Xem mẫu
- om
.c
ng
XỬ LÝ ẢNH
co
an
th
ng
Võ Tuấn Kiệt
o
Bộ môn Viễn thông (112B3)
du
Đại học Bách Khoa TpHCM
u
cu
Email: kietleo@gmail.com
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Chương 7: Chuyên đề xử lý ảnh
Phân lớp
om
Khoảng cách tối thiểu
.c
Lân cận gần nhất (KNN)
ng
Xác suất Bayes
co
an
th
o ng
du
u
cu
2
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Nhận dạng
Pattern: is a arrange of descriptors
Pattern classes: a pattern class is a family of
om
patterns that share some common properties
.c
Pattern recognition/classification: to assign
ng
patterns to their respective classes
co
an
Intelligent
th
Ability to separate relevant information
ng
Ability to learn from examples and to generalize
o
du
knowledge so that it can be used in other
u
situations
cu
Ability to draw conclusions from incomplete
information
3
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Ví dụ
Mấy
om
lớp?
.c
Mấy
ng
co
thông
an
số đo?
th
o ng
du
u
cu
4
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Lý thuyết quyết định
Decision-theoretic approaches to recognition are based on the use
om
decision functions.
.c
Let x ( x , x ,..., x ) represent an n-dimensional pattern vector.
T
1 2 n
ng
For W pattern classes , ,..., , we want to find W decision
1 2 W
co
functions d ( x ), d ( x ),..., d ( x )with the property that, if a pattern x
1 2 W
belongs to class , then
an
i
th
d i (x ) d j (x )
ng j 1 , 2 ,..., W ; j i
The decision boundary separating class
o
and is given by
j
du
i
u
cu
d i (x ) d j (x ) or d i ( x ) d j (x ) 0
5
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Phân lớp khoảng cách tối thiểu
Suppose that we define the prototype of each pattern
class to be the mean vector of the patterns of that class:
om
.c
1
m x
ng
j j
N j x w
where Nj is the number of pattern vectors from class wj j
co
(j=1,2,…,W)
an
th
Using the Euclidean distance to determine closeness
ng
reduces the problem to computing the distance
o
du
measures
u
cu
D j (x) x m j
where T 1/ 2
|| a || (a a )
6
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- We assign x to class wi if Di(x) is the smallest
om
distance.
.c
The smallest distance is equivalent
1
to evaluating the
ng
T T
functions d ( x ) x m m m
j j j j
co
2
an
th
And assign x to class wi if di(x) yields the largest
ng
numerical value.
o
du
The decision boundary between classes and for a
u
minimum distance classifier is
cu
T 1 T
d ij ( x ) d i (x) d j (x) x (m i
m j) (m i
m j ) (m i
m j) 0
2
7
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Giải thuật phân lớp khoảng cách tối thiểu
Each class is represented by its mean vector
om
Training is done using the objects (pixels) of known class
.c
Mean of the feature vectors for the object within the class is
ng
calculated
co
New objects are classified by finding the closest mean vector
an
th
o ng
du
u
cu
8
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Phân lớp lân cận gần nhất
Nearest neighbour classifier: A pattern in the
om
test data is classified by calculating the
.c
distance to all the patterns in the training data.
ng
The class of the training pattern that gives the
co
shortest distance determines the class of the
an
test pattern.
th
o ng
du
u
cu
9
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Giải thuật KNN
om
.c
ng
co
an
th
o ng
du
u
cu
10
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Phân lớp xác suất Bayes
om
.c
ng
co
an
th
o ng
du
Naïve
u
cu
Zero frequency smoothing technique
(Laplace estimation)
11
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Naïve Bayes using m-estimate
om
.c
ng
co
an
th
o ng
du
u
cu
12
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Phân bố xác suất
om
.c
ng
co
an
th
o ng
du
u
cu
13
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Hàm ngẫu nhiên 1 biến
om
.c
ng
co
an
th
o ng
du
u
cu
14
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Hàm ngẫu nhiên 2 biến
om
.c
ng
co
an
Mean
Correlation
th
o ng
du
u
cu
Covariance
15
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Phân bố Gaussian nhiều biến
om
.c
ng
co
an
th
o ng
du
u
cu
16
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- om
.c
ng
co
an
th
o ng
du
u
cu
17
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Ôn tập
Nguyên lý cơ bản của nén ảnh?
om
Các thông số đánh giá hiệu quả nén ảnh?
.c
Các kỹ thuật nén ảnh không tổn hao?
ng
co
Các bước cơ bản trong nén ảnh JPEG?
an
Giải thuật phân lớp khoảng cách tối thiểu?
th
ng
Giải thuật phân lớp lân cận gần nhất?
o
du
Giải thuật phân lớp xác suất Bayes?
u
cu
18
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Bài tập 7
om
.c
ng
co
an
th
o ng
du
u
cu
19
CuuDuongThanCong.com https://fb.com/tailieudientucntt
- Bài tập 8
om
.c
ng
co
an
We are told that the fruit is Long, Sweet and
th
ng
Yellow. Is it a Banana? Is it an Orange? Or is it
o
du
some Other Fruit?
u
cu
20
CuuDuongThanCong.com https://fb.com/tailieudientucntt
nguon tai.lieu . vn