资源简介
在本文中我们展示
在人类视觉中一种有效的色彩外观模型,
其中也包含原则性的参数选择作为一种先天的空间联合机制,可以被推广
以获得优于最新技术的显着性模型楷模。尺度积分是通过逆小波变换实现的
通过一系列比例加权中心环绕响应。比例加权函数(称为ECSF)已被优化以更好地复制心理物理数据颜色的外观,和适当的尺寸中心环绕抑制窗口已被确定通过对眼睛固定数据训练高斯混合模型,从而避免了特别的参数选择。
论文:Saliency Estimation Using a Non-Parametric Low-Level Vision Model
代码片段和文件信息
function [w c] = DWT(image wlev)
% Implementation of Mallate Discrete Wavelet Transform.
%
% outputs:
% w: cell array of length wlev containing wavelet planes in 3
% orientations.
% c: cell array of length c containing residual planes.
%
% inputs:
% image: input image to be decomposed.
% wlev: # of wavelet levels.
% pad image so that dimensions are powers of 2:
image = add_padding(image);
% Defined 1D Gabor-like filter:
h = [1./16.1./4.3./8.1./4.1./16.];
energy = sum(h);
inv_energy = 1/energy;
h = h*inv_energy;
w = cell(wlev1);
c = cell(wlev1);
for s = 1:wlev
img_dim = size(image1);
orig_image = image;
inv_sum = 1/sum(h);
% decimate image along horizontal direction
prod = symmetric_filtering(image h)*inv_sum; % blur
HF = prod;
tmp_prod = zeros(size(prod));
tmp_prod(:1:2:img_dim) = prod(:1:2:img_dim); % downsample
tmp_prod2 = symmetric_filtering(tmp_prod h)*inv_sum; % blur downsampled image horizontally
GF = image - 2*tmp_prod2; % horizontal frequency info
% decimate image along vertical direction
prod = symmetric_filtering(HF h‘)*inv_sum; % blur
HHF = prod;
tmp_prod = zeros(size(prod));
tmp_prod(1:2:img_dim:) = prod(1:2:img_dim:); % downsample
tmp_prod2 = symmetric_filtering(tmp_prod h‘)*inv_sum; % blur downsampled image vertically
GHF = HF - 2*tmp_prod2; % vertical wavelet plane
% decimate GF along vertical direction
prod = symmetric_filtering(GF h‘)*inv_sum; % blur
tmp_prod = zeros(size(prod));
tmp_prod(1:2:img_dim:) = prod(1:2:img_dim:); % downsample
HGF = 2*symmetric_filtering(tmp_prod h‘)*inv_sum; % horizontal wavelet plane
% save horizontal and vertical wavelet planes:
w{s1}(::1) = HGF;
w{s1}(::2) = GHF;
% Downsample residual image HHF:
HHF = HHF(1:2:img_dim1:2:img_dim);
% save residual
C = HHF;
c{s1} = C;
% upsample residual image:
HHF = upsample(upsample(HHF2)‘2)‘;
% blur with vertical filter:
image = 2*symmetric_filtering(HHF h‘)*inv_sum;
% blur with horizontal filter:
image = 2*symmetric_filtering(image h)*inv_sum;
% Create and save wavelet plane:
DF = orig_image - (image + HGF + GHF);
w{s1}(::3) = DF;
% Downsample residual image:
image = HHF(1:2:img_dim1:2:img_dim);
end
end
function image_padded = add_padding(image)
% Pads image so that dimensions are powers of 2.
%
% outputs:
% image_padded: padded image.
%
% inputs:
% image: input image.
[height width] = size(image);
% pad image when dimensions are not powers of 2/equal to each other:
nearest_pow = 2^ceil(log2(max(widthheight)));
image_padded = zeros(nearest_pow);
image_padded(1:height1:width) = image;
image_padd
属性 大小 日期 时间 名称
----------- --------- ---------- ----- ----
....... 65014 2016-01-07 07:59 SIM\3.jpg
....... 36922 2016-01-07 07:59 SIM\35.jpg
....... 3894 2016-01-07 07:59 SIM\DWT.m
....... 1512 2016-01-07 07:59 SIM\generate_csf.m
....... 1001 2016-01-07 07:59 SIM\IDWT.m
....... 1485 2016-01-07 07:59 SIM\README.txt
....... 1064 2016-01-07 07:59 SIM\rgb2opponent.m
....... 3957 2016-01-07 07:59 SIM\SIM.m
....... 730 2016-01-07 07:59 SIM\SIM_demo.m
....... 1050 2016-01-07 07:59 SIM\symmetric_filtering.m
目录 0 2018-05-29 16:42 SIM
----------- --------- ---------- ----- ----
116629 11
- 上一篇:显著性检测SR算法--Matlab
- 下一篇:显著性检测SWD--Matlab
相关资源
- 显著性检测SWD--Matlab
- 显著性检测SR算法--Matlab
- 显著性检测CA算法--Matlab
- 基于matlab的神经网络数字识别
- matlab 信道均衡
- 基于Matlab的语音合成技术
- MIMO的ML检测matlab程序
- MIMO的ZF检测matlab程序
- 多径瑞利信道的matlab仿真程序
- MIMO的OSIC检测算法。基于三种方式。
- matlab实现带权重的线性回归
- matlab交互式图像分割源程序
- 基于 HMM算法的语音识别的matlab程序
- 图像SNR、PSNR与MSE计算
- 2004-c题(饮酒驾车的数学模型)+mat
- 非线性扩散--PM算法完整matlab代码
- MATLAB全规则元胞自动机
- Gabor变换 MATLAB程序 根据理论自己编写
- MatLab 批量将24位PNG彩图转换为8位PNG彩
- 用MATLAB实现的MM1仿真
- matlab 绘图教程
- 超全的模式识别Matlab源程序
- SPWM的三相MMC的matlab仿真
- 谐波检测 matlab程序
- 船舶航迹控制
- 船舶海上运动matlab仿真
- 遗传算法改进的模糊C-均值聚类MATLA
- 二相编码调制脉冲压缩信号仿真
- MATLAB 汽车牌照识别程序
- HVDC MATLAB风电仿真模型
评论
共有 条评论