ﺑﺎزﯾﺎﺑﯽ ﺳﺮﯾﻊ ﺗﺼﺎوﯾﺮ ﺑﺎﻓﺘﯽ ﺑﺎ اﺳﺘﻔﺎده از ﺗﺒﺪﯾﻞ واﻟﺶ- ﻫﺎداﻣﺎرد

Fast Texture Retrieval Using Walsh-Hadamard Transform

ﯾﮑﯽ از ﻣﺴﺎﺋﻞ ﻣﻬﻢ در ﭘﺮدازش ﺗﺼﻮﯾﺮ، ﭘﯿﺪا ﮐﺮدن اﻟﮕﻮرﯾﺘﻢ ﻫﺎﯾﯽ اﺳﺖ ﮐﻪ ﺑﺘﻮان ﺗﺸﺎﺑﻪ و ﻋﺪمﺗﺸﺎﺑﻪ ﯾﮏ ﺗﺼﻮﯾﺮ ﺑﺎﻓﺖ را ﺑﺎ ﺳﺎﯾﺮ ﺗﺼﺎوﯾﺮ در زﻣﺎن ﮐﻮﺗﺎه ﻣﺸﺨﺺ ﮐﻨﺪ. از آﻧﺠﺎﯾﯽ ﮐﻪ ﺗﺼﺎوﯾﺮ ﺑﺎﻓﺖ داراي اﻟﮕﻮﻫﺎي ﺗﮑﺮارﺷﻮﻧﺪه در ﮐﻞ ﺗﺼﻮﯾﺮ ﻫﺴﺘﻨﺪ، اﻟﮕﻮرﯾﺘﻢ ﻫﺎي ﻣﺸﺎﺑﻬﺖﯾﺎﺑﯽ ﺑﺮاي ﺗﺼﺎوﯾﺮ ﻣﻌﻤﻮﻟﯽ ﺑﺮاي ﺗﺼﺎوﯾﺮ ﺑﺎﻓﺖ ﮐﺎراﯾﯽ ﻧﺪارﻧﺪ. اﻟﮕﻮرﯾﺘﻢ ﻫﺎي ﻣﺒﺘﻨﯽ ﺑﺮ ﯾﺎدﮔﯿﺮي ﻋﻤﯿﻖ ﻫﻢ ﻧﯿﺎزﻣﻨﺪ ﺣﺠﻢ ﺑﺴﯿﺎر زﯾﺎدي داده در ﻫﻤﺎن ﮔﺮوه ﻫﺴﺘﻨﺪ و ﺑﺮاي ﺗﺼﺎوﯾﺮ ﺑﺎﻓﺖ ﮐﻪ ﺣﺠﻢ داده زﯾﺎدي در دﺳﺘﺮس ﻧﺒﺎﺷﺪ ﮐﺎراﯾﯽ ﻧﺪارﻧﺪ. در اﯾﻦ ﻣﻘﺎﻟﻪ، اﻟﮕﻮرﯾﺘﻤﯽ ﺑﺮاي ﺟﺴﺘﺠﻮي ﺳﺮﯾﻊ ﺗﺸﺎﺑﻪ ﺑﺎﻓﺘﯽ ﺑﺎ اﺳﺘﻔﺎده از ﺗﺒﺪﯾﻞ واﻟﺶ- ﻫﺎداﻣﺎرد، ﺗﻮﺳﻌﻪ داده ﺷﺪ. اﯾﻦ اﻟﮕﻮرﯾﺘﻢ، از ﺳﻪ ﻣﺮﺣﻠﻪ ﺗﺸﮑﯿﻞ ﺷﺪه اﺳﺖ: در ﻣﺮﺣﻠﻪ اول از ﻓﯿﻠﺘﺮ ﮔﺎﺑﻮر ﺑﺮاي اﺳﺘﺨﺮاج ﺑﺮدارﻫﺎﯾﯽ ﺑﺎ اﺑﻌﺎد ﺑﺎﻻ از ﻫﺮ ﺑﺎﻓﺖ، اﺳﺘﻔﺎده ﺷﺪ. ﺳﭙﺲ، از ﯾﮏ ﺗﺒﺪﯾﻞ ﺗﺼﺎدﻓﯽ ﺷﺪه واﻟﺶ- ﻫﺎداﻣﺎرد اﺳﺘﻔﺎده ﮔﺮدﯾﺪ ﺗﺎ ﺑﺮدارﻫﺎﯾﯽ ﺑﺎ اﺑﻌﺎد ﺑﺎﻻ ﺑﻪ ﺑﺮدارﻫﺎﯾﯽ ﺑﺎ دو ﺑُﻌﺪ ﺟﺎيﮔﺬاري ﺷﻮد. در ﻣﺮﺣﻠﻪ ﺳﻮم، از ﯾﮏ اﻟﮕﻮرﯾﺘﻢ ﺗﻘﺮﯾﺐ ﺑﺮاي اﻧﺪازه ﮔﯿﺮي ﻓﺎﺻﻠﻪ ﺑﯿﻦ دو ﺑﺮدار اﺳﺘﻔﺎده ﺷﺪ ﺗﺎ ﺗﺸﺎﺑﻪ ﯾﺎ ﻋﺪم ﺗﺸﺎﺑﻪ ﺑﯿﻦ دو ﺗﺼﻮﯾﺮ ﺑﺎﻓﺖ ﻣﺸﺨﺺ ﺷﻮد. ﻧﺘﺎﯾﺞ آزﻣﺎﯾﺶ ﻋﻤﻠﯽ اﻟﮕﻮرﯾﺘﻢ ﮔﻮاه اﯾﻦ اﻣﺮ ﻫﺴﺘﻨﺪ ﮐﻪ اﯾﻦ ﺗﻘﺮﯾﺐ ﺑﺮاي ﺑﺎﻓﺖﻫﺎي واﻗﻌﯽ ﻧﺴﺒﺘﺎً ﻣﻨﺎﺳﺐ ﻫﺴﺘﻨﺪ.

One of the most important issues in image processing is to find algorithms that can determine the similarity and dissimilarity of a texture image with other images in a short time. Because texture images have repetitive patterns throughout the image, similarity algorithms for natural or non- texture images are not effective for texture images. Deep learning algorithms also require large amounts of data in the same group and for texture images that do not have much data volumes available, they do not work. In this paper, an algorithm was developed to rapidly search for tissue similarity using the Walsh-Hadamard transform. This algorithm consists of three steps. In the first step, the Gabor filter was used to extract the high-dimensional feature from each texture. Then, a randomized Walsh-Hadamard transform was used to convert high-dimensional feature from each texture into two-dimensional feature. In the third step, an earth mover distance (EMD) approximation algorithm was used to determine the similarity or dissimilarity between two textures that are represented by two-dimensional vectors. The results of the proposed algorithm proved that this approximation algorithm is relatively suitable for real tissues.