After saving your data in an ARFF file format. You need to load it into WEKA. (Make sure you have a string data type.). Then, tokenize your documents using the preferred tokenization method. Once you have the full dictionary, you can apply the stopwords list approach, stemmer, etc.
In general, using WEKA's "StringToWordVector" (unsupervised filter) would help you to go with text analysis, but you need to be carefull, as StringToWordVector is not strictly unsupervised technique when a class attribute is set because it creates a separate dictionary for each class and then merges them.
After filtering the data using the StringToWordVector, you can apply the desired classifier to perform text classification.
If you have an independent test set data, it has to be compatible with the training set that gets loaded from the Preprocess panel.
Let me add some information to clarify certain aspects.
First, the main topic of this post is "text classification" which can be accomplished decently with WEKA. However, image classification, that you highlighted, is not related to this topic. Thus, please next time stick with the topic of the post.
Second, you said " i tried to use it for image classification, but i didn't found any package or library for it......... You git only a single library there which is not enough"
Well, in your statement (up) two contradictory statements where sometimes you say there is no package for image classification, then you declare that got one package. So what's is the finial decision?
Let me tell you that you can apply WEKA through Java (main environment of WEKA and most popular programing language), since you like coding. There you can easilly write a code or perhaps develop any method (in case you don't find a related package in WEKA.) You can also write a Python code through WEKA using the associated wrapper. Finally, R and Matlab can also be applied in WEKA easily.
Kindly in the future, don't judge a tool without having a solid background about it.