What are the common features could be used for those classifiers?

I have built a classifier that either uses K-nearest neighbor or linear discriminant functions to classify objects according to its class by using a training dataset for each class. I have used Maximum length and width, Centroids and medoids for helping the classifier to take the decision. As for an example about extraxting features of the object in the image, supposing that each image has one object with no noise.

for k=1:length(ClassImages)
  ObjectInImage = ExtractObject(image);
  [Medoid, Centroid, Length, Width] = ImageFeatures(ObjectInImage); 
  AllLengths{k} = Length;
  AllWidths{k} = Width;
  Centroids{k} = Centroid;
  Medoids{k} = Medoid;
end

What are another common features could be used for real dataset images, like cars, trees, chairs, and so on?