Innovative Sketch Board Mining for Online image Retrieval

Huda Abdulaali abdul Baqi, Ghazali Sulong, Siti Zaiton Mohd Hashim & Zinah S.Abdul jabar 1 UTM-IRDA Digital Media Centre (MaGIC-X), Faculty of Computing, Universiti Teknologi Malaysia, Malysia 2 UMT. School of Informatics and applied mathematics, Universiti Malaysia Terengganu, 21030 Kuala Terengganu Malysia 3 Big Data Centre, UTM Skudai Johor Takzin, Malaysia 4 University of Al -Mustansiriyah, College of Science, Computer Dep., Baghdad, Iraq


Introduction
Previously, there has been a wide interest and progress on computer aided retrieval of media data.The advances in this area have allowed users to look for a multimedia object in large repositories in a more efficient way.As advances in multimedia retrieval, increase, new interesting, challenging applications are coming up.Image retrieval has very important applications (Banfi, F., 2000) that are beyond the traditional application based on searching such as Architecture and Interior Design, Biochemical, Education,…etc.., for that the easy way to retrieving images is by means textual metadata describing the object in the image.However, then retrieved by text not always come with reliable human tags.Although, many authors have addressed the several methods have been invented (Funkhouser et al., 2003, Eitz, et al., 2012).For example, an easy way to express the user query is by using a line-based hand-draw.One of the current interesting applications is an input draw (sketch) Definition: a sketch consists of a group of elements; a sketch is represented as a structure of a shape used to find images relevant to this shape leading to a method of sketch-based image.
The sketch based image retrieval (SBIR) is part of the image retrieval field.In an SBIR system, the input is a simple sketch representing one or more objects.As shown in Figure 1.
The retriev accurately satisfy a g images fro online ske to the pres features an 2014).In t online.Th semantic s improvem SBIR.Sec Matching.

Related
Although m some relev that an eas the sketch methods.informatio Features ( benchmark user opinio goal of a r procedure into a set geometric enhance th 'keyshape recreate im looks like extract fea al.,2015) i look like a database th enable par representa database, matching m online ima similar im to retrieve way.

Method
The propo and minin finding a      5The equati are: Where X : (1), it will same way.Step 1: Fo

Drawing Question Sketch Online
In this step, the method will be used for implementation.Where queries (drawings of a sketch by user) are entered and the attributes of the drawn sketch are extracted to create a vector of words (attribute), this vector will be compared with each template saved in mysql data base to match with suitable template in the database.This is the first function of to finding the mean of the input sketch and is accomplished by implementing the following: Step 1: input queries (new sketch) Step2: perform algorithm (1 to 2) to extract the keyshape geometric elements of sketch and make it a vector of words (attribute) as an input sketch.
Step3: match the template vector with all templates saved inMysql database.
Step 4: find the most similar template in the database.
Step 5: capture the metadata of the similar template with query template.

Semantic Matching
Semantic matching is a technique used in computer science to identify the information which is semantically correlated (Chauhan et al., 2013), In this process the exact match retrieval techniques provide a basic token matching capability in that only the documents that exactly match with the specified query can be retrieved consisting of one word or several words, considered as an abstract unit, and applied to a family of words related by meaning.

Similarity Matching
Matching is a technique used in computer science to identify the information that is a similarity correlated (Aziz and Rafi, 2010).In this step, the exact match retrieval technique that provides a basic matching capability is the document image metadata annotation that exactly matches the specified query that can be retrieved, consisting of one or several words.
Many methods have been proposed to match the sketch with images in dataset by using the low-level features process in order to make the image look like a sketch and match it with the sketch to be retrieved.This method has a good result when dealing with a small or limited database and retrieved images offline, but is not suitable for large database (Parui and Mittal, 2015;Bozas, 2012).
In this study, a new approach has been proposed to match the images with sketch in a high-level domain.That means understanding and recognizing the sketch meaning by analysing the geometrical function of the sketch, in order to return images based on that.In this step, it is important to mention that the matching was done using the process is presented in two tasks as below: 1. Linear pattern Similarity matching: this matching is performed between the vector of word (template) coming from users with database templates to find more vectors similar to it.
2. Matching with Google Engine: this matching is performed between the metadata captured from the (query) and submitted by URL engrained to be matched with the metadata annotation in Google images.
Linear pattern Similarity matching.It is important to clarify that mathematical similarity approach (Cosine similarity based on dot product) will be used to match the template of input vector with all templates saved database and to get the metadata related to input query.
The linear pattern similarity matching is used for matching the template of elements that came from the query with a template(s).Here, the cosine linear algorithm is used to measure the distance similarity between the template (query) with each template saved in mysql of two vectors (query and template stored).Step 1: Input: users query ( concept) Step 2: go search connection as URL connection to Google URL and user Query; Step 3: set go search Connection Properties as Method = 'send'; User-Agent =' Google Chrome'; Web sockets = search Connection.Open; Step 4: Get an input stream from send concept to stream reader of search engine ; While stream reader has imgurl do Add current imgurl to imagurl_list; Step 5: return imgurl-list;

Results and Discussion
The sketch is conceptualized and used to retrieve the images from the web, which is based on the image automatic web annotation (Google images, flicker).Ten free hand sketches are used to retrieve the images from the web.Our method is tested by using the first page contain 30 images appeared on the search of each sketch, which are drawn on the sketch board online.The annotation features are used to acquire all the information, which can be used as a matching tool for the query to create an ontological form representing the sketch concept.
The performance of the proposed method is illustrated in Fig. 15.The top five retrieved images for six input sketches are shown.The precision is evaluated by recalling the corresponding images for each sketch after retrieval.The sketch is conceptualized and a ranking score calculated for each image by precision@10 and achieved rate within 0.87 to 0.99.Fig. 16, shows the precision-recall graphic compared to BoF, Key shape-based and Keyshape+BoF approaches (Saavedra, and Bustos, 2014).This evaluation shows that the performance of the keyshape mining based proposal increases the retrieval effectiveness, which satisfy correctly identified images True Positive (TP) for accuracy of retrieved images.This implies that all the images are retrieved.Moreover, the sketch specification is easy for transforming into the general ontological concept in a way that reduces False Negative (FN) images.To provide easy comparisons on a standard dataset and compute the recall which is difficult for a large dataset, we tested our system by using the same sketch shapes belonging to (Eitz et al., 2011).The Mean Average Precision (MAP) compared with the key shape base achieved (0.85639) where (keys-shape +BOF) achieved (0.88722) (Saavedra, and Bustos, 2014), and our method of mining the keyshape achieved (0.90422).Our proposed system not only performed accurately, but time is not a concern in retrieving depending on the network connection speed.
User dissatisfaction appears from the simple geometrical shape (set of lines, circle, or ellipse) used to draw the sketch.However, it is evident from the figure that when the shape elements become more complex the recall becomes higher and the precision is reduced.This results in the usual response in terms of the appearance of many unwanted images.However, the retrieved images are more specific.
The sketch board has a set of attributes with specific types.The nature of each attribute can be defined through the sketch board edition.To create a new sketch board, the following types of attributes are activated to display a plain hand draw type and a sketch category value from a range of predefined values of sketch elements.Edit the category records by clicking on the corresponding cell on sketch board.Query uploads can display the retrieved image path directly from the web and edits multi-category records on the sketch board.

Figure
Figure 2. G

Figure
Figure 5. T northward Figu Figure Figure 1

Figure
Figure 13.Ex Figure 15 hand draw retrieval on

Table 1 .
Explain the templates related to two sketches, moon and pyramid