Learning Feature Representations with K-Means

Hand Vein Recognition with Single-layer Feature Learning Model

“…K-means has already been identified as a successful method to learn features from images by computer vision researchers. The popular "bag of features" model [15] [16] from the computer vision community is very similar to the pipeline that we will use in this chapter, and many conclusions here are similar to those identified by vision researchers [17] [18]. So the paper introduces the K-means to realize the vein feature representation system so as to find the feature distribution without adopting hand-crafted feature as the prior knowledge, and the experimental results demonstrate the efficiency of the proposed single-layer feature learning model in solving the feature learning problem in hand vein recognition task.…”

“…The expression of stands for the code vector of the input vector .The first constrain is set to guarantee the sparsity with the condition that there exists one nonzero item at most, while the second criterion is set to avoid that there exists some unexpected data with large value in the feature vectors, which is the set to realize the constrain on the dictionary, the specific solving process is according to the methods in [18]. The final process is coding on the new input vectors to get the features after getting the training dictionary:…”

Hand Vein Image Enhancement Based on Multi-Scale Top-Hat Transform

“…For future extension of our work, we are going to work on designing an unsupervised method (i.e., k-means [13] and CNN [14]) to figure out the better control factor in edge gradient information calculation. What's more, direct enhancement on RGB images will be tried and discussed later to find another efficient framework on not only vein images but also other low-contrast medical images enhancement problem.…”

“…Suppose that the maximum decomposition scale is set as n, the result after multiscale top-hat processing based on the above equations could be expressed as (12) Xop → {Xop o1 , Xop 2 , ⋯ , Xop ⋯ , Xop }, (13) Xcl → {Xcl 1 , Xcl 2 , ⋯ , Xcl ⋯ , Xcl }. Xop , Xcl represent the set of detailed information at the adjacent scale, and the key design for the later enhancement process is how to effectively enhance the detailed information while avoiding importing fake vein information.…”

Deep Learning for Intelligent Substation Device Infrared Fault Image Analysis

“…As mentioned in [18] and [19], they found out that K-means algorithm attains best performance with only one parameter to tune. On the other hand, [20] proposed a new feature learning algorithm called sparse filtering.…”