**6. Usability testing**

In order to infer the usability of the proposed CAPTCHA, two users' acceptance tests have been conducted on this technique. The first test was aimed at taking a large number of responses from different users, while the second test was aimed at testing a large number of the produced CAPTCHA images.

**6.2. Second usability test**

CAPTCHA correctly or not.

answers were recorded.

**7. Conclusion**

applications.

CAPTCHA samples [1].

each image were misinterpreted by the users.

lected handwritten characters database [1].

The second usability test was carried out using the implemented web pages to give users a real experience similar to solving a real CAPTCHA on any website. Therefore, as mentioned earlier, the implemented web pages produce CAPTCHA images at the real time and responds to users as soon as they click on the "validate" button to inform them either they solve the

Innovative Multilingual CAPTCHA Based on Handwritten Characteristics

http://dx.doi.org/10.5772/intechopen.72599

173

Consequently, the test was conducted on 20 users with five CAPTCHA images average for each user; each user viewed the CAPTCHA on the web page and solved five different CAPTCHAs, and in the meanwhile the users' answers and the web page responses to their

Correspondingly, the testing results showed that 92% of the tested images were correctly recognized by users, while in the remaining 8% of the images fewer than three characters of

However, the percentages of the results have proven a fair usability percentage, which also could be further improved with little adjustments on the distortion methods and on the col-

The rapid evolution of web and mobile applications turn these applications into an important part in people's daily life, where people rely on them to accomplish most of their activities. On the other hand, all the rapid improvement in these applications comes with a rapid increase in the number of malicious bots and applications that threatens the security of web and mobile

Therefore, we introduced a new CAPTCHA technique that utilizes handwriting styles and we have put it through several experiments to adjust, improve, and test the technique while

The introduced technique is a novel handwritten CAPTCHA, which basically relies on employing the handwriting characteristics that can only be interpreted by humans while being comparatively hard for OCRs to recognize. The proposed approach adopts four different languages (English, Arabic, Spanish, and French); each language has its own handwritten

Moreover, few testing experiments have been conducted on the proposed CAPTCHA to test its robustness as well as the level of security it provides. The experiments were done using six different OCRs on 500 different CAPTCHA samples. Nevertheless, the results of the experiments manifest the significant benefits of utilizing handwriting characteristics with

trying to reflect every needed adjustment to the technique immediately.

characters used in synthesizing the CAPTCHA text.

#### **6.1. First usability test**

The first test targeted 100 users through an online survey that aimed at understanding how users will interpret five different CAPTCHA images. These five images were chosen on the basis of different aspects to study users' responses regarding characters' distortions and unclear handwriting styles (see Appendix B for CAPTCHA images used in the survey).

**Table 2** illustrates the results of the conducted survey and shows the answers' patterns for each CAPTCHA image used in the survey.

According to the results shown in **Table 2**, 82% of the users were able to correctly recognize the CAPTCHA characters of the first image, the remaining 18% failed and were confused between characters, and noise, and distortion.

As for the second and third images, they were recognized by 85 and 75% of the users, respectively, while the rest of the users were confounded by one character, due to the warping distortion method applied on that character.

Moreover, users succeed in correctly recognizing the fourth CAPTCHA image with a percentage of 74%, while 61% of them correctly recognized the fourth and fifth CAPTCHA images. However, the remain percentages of the users who failed to correctly recognize the last two CAPTCHA images have failed because of the unclear handwriting style which was selected on purpose to reflect the worst cases that could be produced from the collected database.

In general, it must be mentioned that the partially correctly recognized pattern indicates that the user misinterpreted three characters or fewer from the CAPTCHA word, otherwise it will be considered that the user has incorrectly recognized the whole CAPTCHA word.

Additionally, the average time taken by each participant to solve the survey was 2 minutes and 38 seconds.


**Table 2.** Results of the survey conducted for the first technique.

#### **6.2. Second usability test**

**6. Usability testing**

172 Multilingualism and Bilingualism

**6.1. First usability test**

and 38 seconds.

the produced CAPTCHA images.

each CAPTCHA image used in the survey.

between characters, and noise, and distortion.

**Table 2.** Results of the survey conducted for the first technique.

tortion method applied on that character.

In order to infer the usability of the proposed CAPTCHA, two users' acceptance tests have been conducted on this technique. The first test was aimed at taking a large number of responses from different users, while the second test was aimed at testing a large number of

The first test targeted 100 users through an online survey that aimed at understanding how users will interpret five different CAPTCHA images. These five images were chosen on the basis of different aspects to study users' responses regarding characters' distortions and unclear handwriting styles (see Appendix B for CAPTCHA images used in the survey).

**Table 2** illustrates the results of the conducted survey and shows the answers' patterns for

According to the results shown in **Table 2**, 82% of the users were able to correctly recognize the CAPTCHA characters of the first image, the remaining 18% failed and were confused

As for the second and third images, they were recognized by 85 and 75% of the users, respectively, while the rest of the users were confounded by one character, due to the warping dis-

Moreover, users succeed in correctly recognizing the fourth CAPTCHA image with a percentage of 74%, while 61% of them correctly recognized the fourth and fifth CAPTCHA images. However, the remain percentages of the users who failed to correctly recognize the last two CAPTCHA images have failed because of the unclear handwriting style which was selected on purpose to reflect the worst cases that could be produced from the collected database.

In general, it must be mentioned that the partially correctly recognized pattern indicates that the user misinterpreted three characters or fewer from the CAPTCHA word, otherwise it will

Additionally, the average time taken by each participant to solve the survey was 2 minutes

**Pattern First image Second image Third image Fourth image Fifth image**

be considered that the user has incorrectly recognized the whole CAPTCHA word.

Correctly recognized 82 85 75 74 61 Partially correctly recognized 18 15 25 26 39 Incorrectly recognized 0 0 0 0 0

The second usability test was carried out using the implemented web pages to give users a real experience similar to solving a real CAPTCHA on any website. Therefore, as mentioned earlier, the implemented web pages produce CAPTCHA images at the real time and responds to users as soon as they click on the "validate" button to inform them either they solve the CAPTCHA correctly or not.

Consequently, the test was conducted on 20 users with five CAPTCHA images average for each user; each user viewed the CAPTCHA on the web page and solved five different CAPTCHAs, and in the meanwhile the users' answers and the web page responses to their answers were recorded.

Correspondingly, the testing results showed that 92% of the tested images were correctly recognized by users, while in the remaining 8% of the images fewer than three characters of each image were misinterpreted by the users.

However, the percentages of the results have proven a fair usability percentage, which also could be further improved with little adjustments on the distortion methods and on the collected handwritten characters database [1].
