Here is the ascii file of my last version of features, I have added fourier transforms, radons and dct similar to Dr Katanforoush’s voice.
in order to use the features load the file in matlab to use.
The set of 29 Features include: number of components, glcm contrast,image momentums(mean, std,skewness,kurtosis, uniformity,entropy),3first columns of 3rd fourier(row-column-column), radons of 3rd fourier in 4 angles(each covolved in 5 bin histogram), tamura texture coarseness.From: Mina Naghshnejad <mina.naghshne...@gmail.com>
Date: Tue, 3 Jan 2012 22:32:13 -0800 (PST)
Local: Wed, Jan 4 2012 9:32 am
Subject: Re: latest version of features
Reply | Reply to author | Forward | Print | Individual message | Show
original | Remove | Report this message | Find messages by this
author
سلام زهرا جان.
روش های مختلفی برای استخراج فیچر از تبدیل ها وجود داره. مثل هیستوگرام
گرفتن از کل ماتریس و یا انتخاب چند ستون ( ستون های اول) یا سطر از
تبدیل هایی که روی تصویر زده شده. این روش ها تست شدن و هر کدوم که
نتیجه ی بهتری داد در نهایت مورد استفاده قرار گرفت اما این فیچرها
فیچرهای نهایی ما نیستند و مسلما قراره بهبود پیدا کنند فقط برای اینکه
دوستانی که روی قسمت
learning
کار می کنند داده برای کار داشته باشند به اشتراک گذاشته شده اند .
درنهایت
پیشنهاد می کنم از اونجایی که دوهفته بیشتر تا امتحان فرصت نداریم جلسه
ی
گروه رو در بعد از ظهر روز امتحان برگزاری کنیم و در اونجا می تونیم به
تفضیل در این موارد بحث کنیم ..
متشکرم
نقش نژاد
On Jan 3, 10:44 pm, zahra rezvani <rezvaniza...@gmail.com> wrote:
> سلام مینا جان
> از این تبدیل ها چه طور این
> feature
> ها را به دست آوردی ؟
>
> با تشکر
> زهرا رضوانی
> On 3 January 2012 11:07, Mina Naghshnejad <mina.naghshne...@gmail.com>wrote:
>
>
>
> > Here is the ascii file of my last version of features, I have added
> > fourier transforms, radons and dct similar to Dr Katanforoush’s voice.
>
> > in order to use the features load the file in matlab to use.
> > The set of 29 Features include: number of components, glcm contrast,image
> > momentums(mean, std,skewness,kurtosis, uniformity,entropy),3first columns
> > of 3rd fourier(row-column-column), radons of 3rd fourier in 4 angles(each
> > covolved in 5 bin histogram), tamura texture coarseness.
> > the last column is the target classes of data(1to6).
> > the best score I reached in cross validation with selecting randomly 0.7
> > of data set as training set with Matllab K nearest neighbor classifier with
> > K=8 was 0.72.
>
> > Regards,
> > Mina Naghshnejad- Hide quoted text -
>
> - Show quoted text -