We revisit the problem with generic object recognition from the point of view of human-computer interaction. While many existing algorithms for generic object recognition first try to detect target objects before features are extracted and classified in processing, our work is motivated by the belief that solving the task of detection by computer is not always necessary in many practical situations, such as those involving mobile recognition systems with touch displays and cameras. It is natural for these systems to ask users to input the segmentation data for targets through their touch displays. Speaking from the perspective of usability, such systems should involve rough segmentation to reduce the user workload. In this situation, different people would provide different segmentation data. Here, an interesting question arises – if multiple training samples are generated from a single image by using various segmentation data created by different people, what would happen to the accuracy of classification? We created “20 wild bird datasets” that had a large number of rough segmentation datasets made by 383 people in an attempt to answer this question. Our experiments revealed two interesting facts: (i) generating multiple training samples from a single image had positive effects on classification accuracies, especially when image features including spatial information were used and (ii) augmenting training samples with artificial segmentation data synthesized with a morphing technique also had slightly positive effects on classification accuracies.

Histogram-based image features such as HoG, SIFT and histogram of visual words are generally represented as high-dimensional, non-negative vectors. We propose a supervised method of reducing the dimensionality of histogram-based features by using non-negative matrix factorization (NMF). We define a cost function for supervised NMF that consists of two terms. The first term is the generalized divergence term between an input matrix and a product of factorized matrices. The second term is the penalty term that reflects prior knowledge on a training set by assigning predefined constants to cannot-links and must-links in pairs of training data. A multiplicative update rule for minimizing the newly-defined cost function is also proposed. We tested our method on a task of scene classification using histograms of visual words. The experimental results revealed that each of the low-dimensional basis vectors obtained from the proposed method only appeared in a single specific category in most cases. This interesting characteristic not only makes it easy to interpret the meaning of each basis but also improves the power of classification.