Software Usability Research Of Verific Ation Code Based On ELM Algorithm

With the rapidly development of Internet, the network technology is gradually flooded into our life and work, and brings the issue of network security, Verification code mechanism came into being. The verification code is a kind of effective mechanism to maintain the network secur ity. Verification code is an effective and widely used solution for preventing computer programs(bots) from performing automated but often malicious actions, but it appears a certain influence on site the user of the system. To make verification code robust to automatic character recognition techniques, the texts in the test are often distorted, blurred, and obscure. Therefore, we are facing a dilemma, that is, a verification code should be robust enough so that it can't be broken by programs, but also needs to be meet the requirements of the human-computer interaction so that users need not to verify many times because of the guesses wrong.This paper mainly starts from the perspective of human interaction requirements, selects four kinds of the text verification code with typical representative, and analyzes their software usability. Experiment is recruiting a group of some sighted users participate in the verification code artificial identification, another set of experiments is the verification code automatic identification based on ELM algorithm, and the results of two sets of experiments for the usability analysis, it is concluded that the software usability classification of the verification code.In this paper, the main work is as follows:1. In view of the existing verification code samples, artificial identification is carried out under normal condition, and the results of the analysis and statistics are analyzed, which is including the correct identification, the error identification, the easy identification and the difficult identification code data.2. Automatic identification verification code samples. Due to the different characteristics of verification code, so this paper adopts different preprocessing operations to the verification code sample set, then takes character segmentation to the result image by combined with the total number of character in verification code or width constraints based on contour projection method, and then extracts features of a single character, constructs classifier based ELM algorithms trained to recognize. Finally the corresponding classifier is adopted to identify the captcha test sample.3. Analyze the experimental results of artificial recognition and automatic recognition from usability analysis of the verification code. Artificial recognition results show the effectiveness and subjective satisfaction, which can be divided into two types, easy to identify and hard to identify; Automatic identification results show that the robustness of verification code, which can be divided into the machines easy to identify and difficult to identify the two types. Recognition rate is the indicator to distinguish easily recognizable and difficult to identify, and recognition rate more than 70% is easy to recognize, recognition rate lower than 40% is difficult to identify, observe the results of two experiments, four types are divided according to human easy recognition, machine easy recognition, human difficult to identify, machine difficult to identify.