An Efficient Feature Extraction Approach For Pill Identification using Image Mining Techniques

hema annaji rao manikandan

Abstract


Abstract—With the help of image mining techniques, an automatic pill identification system was investigated in this study for matching the images of the pills based on its several features like imprint, color, size and shape. Image mining is an inter-disciplinary task requiring expertise from various fields such as computer vision, image retrieval, image matching and pattern recognition. Image mining is the method in which the unusual patterns are detected so that both hidden and useful data images can only be stored in large database. It involves two different approaches for image matching. This research presents a drug identification, registration, detection and matching, Text, color and shape extraction of the image with image mining concept to identify the legal and illegal pills with more accuracy. Initially, the preprocessing process is carried out using novel interpolation algorithm. The main aim of this interpolation algorithm is to reduce the artifacts, blurring and jagged edges introduced during up-sampling. After preprocessing of the image samples, image registration is proposed to register the high frequency pixels of the interpolated image samples. Then in text and shape features are extracted using Geometrical Gradient feature transformation algorithm. In text feature extraction, initially, only foreground of the pill image is considered by removing background pixels. Also, color extraction is carried out using color histogram approach. Finally, Feature Matching is done using Cross Correlation approach, in which the resultant value in the vector form of the query image is compared with the resultant value of the interpolated database image using cross correlation method. If the vector values of the both images are matched then the pill is considered as legal else it is declared as an illegal pill. Simulation results shows that the proposed techniques provide accurate retrieval results both in terms of time and accuracy when compared to conventional approaches. 


Full Text:

PDF

Refbacks

  • There are currently no refbacks.


Subscribe to Print Journals

 IJAIKD is currently Indexed By   


 http://rgjournals.com/public/site/images/mittalberi/scholar_logo_lg_2011.gif  Journal Seek